Compare commits

...
63 Commits
Author SHA1 Message Date
sam 0fe3caa175 chore(release): v1.46.0
deploy / gate (push) Failing after 6m5s
deploy / build-and-push (push) Skipped
deploy / dist-zip (push) Skipped
2026-10-02 13:03:00 +10:00
sam 83d57ec6a2 fix(daemon): drop the import and assignment left by the composition cleanup
l45's refactor removed the assigned-but-unread daemon handles and l44 had
already moved the guardrail wiring out of the daemon, so the rebase union
left internal/daemon/daemon.go importing internal/tools with no use and
internal/app/archied/bootstrap.go assigning a Guardrails field that no
longer exists. main did not build.

Caught by building the merged tree rather than trusting the merge; the
merge train merged on a failed gate instead of chaining on it, which is
the mistake this commit repairs.
2026-10-02 13:00:56 +10:00
sam e762f15844 docs(agent-system): failure classification lives on the persisted record
The hazards list still claimed failure classification and retryability are
not represented explicitly. The persisted execution record has carried the
park class the parking code records since the typed-park work landed, so
the hazard now points at that record and the taskstate vocabulary instead
of restating a field list.

Regenerated the committed development-docs artifact for the edited page.
2026-10-02 12:57:53 +10:00
sam 1539fdc560 fix(daemon): classify workflow-pin parks by cause, not a blanket transient
pinTaskProfile parked every pinWorkflowDefinition failure as
taskstate.ParkTransient. That class documents that a requeue can succeed,
but the dominant pin failures are durable: a stored definition whose YAML
does not hash to the digest beside it, a definition that does not parse,
and a workflow or playbook binding resolving to an undefined workflow all
recur on every requeue until an operator repairs the row or the
configuration. A future retry pass reading transient as its input would
loop on them forever.

pinFailure now carries the class its cause needs, decided where the cause
is known, and pinParkClass reads it at the park site. A durable pin defect
parks operator-actionable; only a State Store read or persist failure --
the genuinely environmental case -- stays transient. The safe default for
an unclassified cause is ParkNeedsHuman, never a requeue that will fix it.

Red-first: TestDurablePinDefectIsNotParkedTransient failed at its
assertion before the fix (parked "transient"); TestTransientPinFailureStaysTransient
keeps the environmental case transient so the rule is per-cause, not a
blanket flip.
2026-10-02 12:57:53 +10:00
sam 0e37a0f536 refactor(daemon): drop assigned-but-unread composition handles
Three exported Daemon fields were wired by the composition root and read
nowhere: CapabilityHost (*plugin.Host), Curators (*curator.Registry) and
ToolRegistry (*tools.Registry). No field access exists anywhere in the
repo, and git history shows no read ever existed.

Each handle has a live owner outside the daemon: the curator
registry/runtime is boot-owned (the curator isolation invariant keeps it
off the agent loop), the capability host is lifecycle-owned by the
composition root and consumed by the control plane, and the tool registry
is consumed by chat, curator and MCP wiring. Keeping the daemon copies was
decorative wiring, so the fields, their composition assignments and the
now-unused plugin/curator imports go together.

The daemon architecture test now refuses the three names, and the plugin
host contract test refuses a CapabilityHost: handle on the daemon alongside
the PluginRegistry: one it already refused.
2026-10-02 12:57:53 +10:00
sam de32542c00 fix(agentworker): supply the guardrail engine to every task run
TaskContext.Guardrails was never assigned: the engine was built in
archied's composition root and held on Daemon.Guardrails, which no code
read after workflow execution moved to the full-task worker handoff.
archie-agent is a separate process, so a live engine cannot cross that
handoff. Build the engine at the agentworker composition root (the
process that runs agent stages), thread it through the task dependencies,
and set it on TaskContext; drop the now-dead archied-side construction.

Refs: archie-core-jd5v
2026-10-02 12:37:35 +10:00
sam c1673e43fb fix(config): carry max_retries into the worker TaskConfig snapshot
The remediate round cap runs inside archie-agent and reads
TaskContext.Cfg.MaxRetries, but TaskConfig did not carry the global
Config.MaxRetries: ForTask and ToConfig both dropped it, so a default
configuration reached the cap as 0 and a remediation never parked. Carry
the value (a repo-level override still wins via EffectiveMaxRetries) and
pin it with a red-first cap test traced through ForTask/ToConfig.

Refs: archie-core-ad3p
2026-10-02 12:37:35 +10:00
sam a9233755c3 fix(workflow): bind the remediate worktree before the builder runs
Commit 4e5b91d7 moved the remediate resume into daemon worktree
preparation and deleted StageResumeWorktree together with the Dir/Resume
pair it called, but nothing in the workflow replaced the binding it did:
TaskContext.Dir and Branch stayed empty for a whole run, so
remediate-build ran the builder with no workspace and
remediate-commit-push called CommitAll/Push with an empty dir and branch.

Extend the workflow-side Trees seam with the prepare target, mirroring
the daemon's mode-aware worktree.Manager.Prepare: PrepareFresh lands on
base, a non-empty target is the branch a resume continues. Remediate now
stages StagePrepareWorktreeOnBranch before any agent stage, binding the
same directory the daemon prepared and the task's PR branch, and failing
closed when the task has no branch rather than preparing base. The
container adapter returns the requested resume target verbatim.

Refs archie-core-h4tk
2026-10-02 12:29:57 +10:00
sam 7ca81dc85a feat(agentworker): answer MCP sampling from the task's model
A task hosts the same configured MCP servers as a chat turn, but the
worker's provider builder passed no sampling handler, so a task-scoped
server's sampling/createMessage request got a JSON-RPC method-not-found
while the identical request succeeded on the chat path. The worker runs
its own model runtime, so it can answer.

Thread a sampling handler built from the task's builder model into the
worker's MCP provider builder, beside the chat path's handler. Build the
task's model runtime once and share it with the stage runners instead of
creating a second client for sampling.

Move the sampling message mapping and max-token bound to agentexec, next
to the runtime both paths call models through, so a server gets the same
bound on either path. Carry the catalog's reasoning flag in
ModelLimits: a reasoning-class builder model must omit the max-token
bound its provider rejects, exactly as the chat path does.
2026-10-02 12:20:31 +10:00
sam 457da2e97d refactor(agent): drop the SOUL upgrade branch with no shipped template
ShippedSoul returned an empty SoulDocument.Legacy, so SoulSuperseded,
replaceSoul and SoulUpgraded had no production producer: the only body the
build ships is its current default, which classifies as SoulCurrent, and only
tests constructed a Legacy document. No earlier SOUL ever shipped (the file was
introduced with the starter in cc83c9ae), so the upgrade path has nothing to
consume.

Remove the speculative machinery and keep the behaviour that is real: seed the
starter when no file exists, recognise the shipped default as current, and
preserve every other body, including a template-shaped edit. A build that
supersedes the starter reintroduces the upgrade together with the old body it
consumes.
2026-10-02 11:59:54 +10:00
sam 656d21520f fix(webui): pin the dashboard validators to the control plane's
ui/src/settings/validation.ts hand-mirrors three control-plane validators with
no shared source, and the model-role rule had already diverged: the dashboard
required a stricter regex while the server only asks for a slash, so it refused
values the server accepts.

Add a shared fixture (internal/app/controlplane/testdata/dashboard-validators.json)
that both sides run through their real implementations: a Go test over the
server validators and ui/test/control-plane-validator-parity.test.ts over the
dashboard ones. Either side changing a rule alone now fails the other's test.

Fix the diverged rule to mirror the server's predicate exactly (non-blank role,
model containing a slash), so an empty or whitespace-padded model is judged the
same way the control plane judges it. Regenerate ui/dist and the dev-docs
artifact.
2026-10-02 11:59:17 +10:00
sam 92eabbed17 feat(telegram): carry clarify and picker through the text fallback
Telegram had no generic question or choice widget in the agent-turn path,
so the interactive contract could not be carried by the only channel that
ships. telegramInteractor implements ClarifyRequester and PickerRequester
on the shared TextFallback and reports both capabilities honestly: the
capability is about carrying the interaction, not the widget, and the cost
is a typed answer. A reply typed while a prompt is pending is consumed by
the blocked interaction rather than dispatched as a new turn; a command is
never consumed, so /stop still cancels the turn.

The carrier is set on the turn context in the Messaging process. It does
not cross the gRPC boundary into the Gateway process, so in the extracted
messaging deployment the gateway turn has no interactive adapter and
ask_user is not offered -- the same gap that already makes
WithApprovalRequester unreachable there. That transport is a product
decision recorded on the lane, not silently worked around; the in-process
path, the capability report and the tests land here.

Red-first: TestInteractivePickUsesTextFallback and
TestInteractiveClarifyReturnsTypedAnswer (plus the command guard) failed at
the assertion before the interactor was implemented.
2026-10-02 11:52:07 +10:00
sam 28acf2a8db feat(gateway): offer ask_user when the turn's channel can carry a question
prepareTurn now registers an ask_user tool only when the turn context
carries an interactive adapter (WithInteractive), so a contract with no
caller has a real one and a channel that cannot ask registers no tool
rather than one that always fails. A channel reporting Picker gets the
native single-choice request; one reporting only Clarify still carries a
choice, rendered as numbered text and matched from the typed reply through
the domain's FormatPickerText/ParsePickerReply rather than erroring.

Red-first: InteractiveTools returned nothing and
TestInteractiveToolsRegisteredWithCarrier / TestAskUserUsesPickerWhenChannelCanPick
/ TestAskUserFallsBackToTextWhenChannelCannotPick failed at the assertion
before the tool was implemented. TestTurnRegistersAskUserOnlyWithCarrier
proves reachability from a real TurnRunner turn.
2026-10-02 11:52:07 +10:00
sam c02852d87f feat(messaging): add a per-turn interactive carrier for clarify and picker
The contract could carry a question but nothing carried the ability to ask
one. Add Interactive, resolved from a channel adapter's own capability
report through InteractiveOf, with WithInteractive / InteractiveFromContext
as the per-turn carrier -- the same shape WithApprovalRequester already uses
for consent. The capability still wins over the method set, so a
mis-reporting adapter contributes nothing instead of being called.

Tests: TestInteractiveOfTrustsCapabilityReport and
TestInteractiveCarrierRoundTrip pin the resolution and the round trip.
2026-10-02 11:52:06 +10:00
sam ab86442156 feat(tools): reconcile MCP servers and tool settings live
Flip tool-settings to ApplyMode live (archie-core-zfb0.3): the daemon now
watches the kind, diffs the stored MCP server set against the running
providers -- connecting added servers, disconnecting removed ones and
reconnecting changed ones while an unchanged server keeps running -- and
rebuilds the web_fetch and minimax entries, whose config is captured at
construction. A changed server whose replacement fails to start is rolled
back to the one running, and the refusal is reported through apply status.

toolprovider.Registry grows running-state Add/Remove/Replace beside
Register, Start and Stop, isolating a provider's failure the way the plugin
host's Add/Replace do. The chat turn model resolves its tool-output policy
per turn instead of capturing it at construction, and the boot registration
reads the running holder so a stored update that lands during boot is not
lost.
2026-10-02 11:46:37 +10:00
sam bfb909a93a docs(prreview): the approval path keeps the task's workflow, it does not force implement
The PRD still described both approval surfaces requeueing under a hardcoded
implement workflow. The shipped path passes an empty workflow, which Requeue
reads as keep-the-task's, so the feasibility handoff has to name implement on
the task before the wait rather than relying on the handler.
2026-10-02 11:45:46 +10:00
sam 894fffc63b test(messaging): widen the config-read gate past explicit config.Config types
The gate that keeps a PRD-forbidden [models]/[providers] read out of the
Messaging Service only blamed identifiers whose parameter or var type spelled
config.Config. A value reached through an inferred declaration
("cfg := doc.Config") or held in a struct field therefore left the read
intact while the rule saw no holder -- the same class of miss that let voice
transcription ship behind a green link gate.

Widen the holder detection to track the config type through inferred short
variable declarations and vars, struct fields and receivers, type-preserving
expressions, and range variables over a slice or map of config, then assert
each previously-inert shape red first.
2026-10-02 11:44:51 +10:00
sam c94be59952 feat(container): apply container-runtime-policies live
The container runtime policies kind was restart-required because the pool
captured image, pull policy, network, max uptime and max concurrency at
construction and the dispatcher fixed its slot count at the start of a drain
pass. Both consumers now re-read the published config:

- the pool's settings come from a live source read per acquire, with the
  boot-resolved registry credential and Docker construction values preserved;
  a live default-image change is pulled rather than assumed present;
- the daemon owns one long-lived dispatcher that accepts a new limit: running
  tasks keep their slots, a raise admits queued work immediately, and a lower
  limit is applied as tasks finish;
- the control-plane descriptor is live, the kind joins runtimeResourceKinds,
  and the apply/reload paths resize the dispatcher after publishing. The
  ContainerConfig fields whose consumers now re-read are added to the reload
  allowlist; registry_auth stays restart-required.

Regenerate the docs artifacts. contracts.json also carries the voice
media-type description a prior commit left stale.
2026-10-02 11:39:46 +10:00
sam 04d0495f12 revert(tools): drop the checkpoint writer with no restore caller
The pre-mutation file checkpoint added in e10d862f wrote a full copy of
every write/edit target into .git/archie-checkpoints, but nothing could
restore it: CheckpointStore.RestoreLatest (and its latest/restore
helpers) had no production caller, no rollback surface reaches the
workspace, and the store was never pruned, so it grew without bound in
every workspace.

A writer whose only reader is unreachable is dead weight and an
unbounded disk cost, so the write half goes with the read half: write
and edit no longer snapshot, and CheckpointStore is deleted. The
regression test pins the decision by asserting a write and an edit
leave no .git/archie-checkpoints behind.
2026-10-02 11:36:10 +10:00
sam 9ccb6e2418 feat(messaging): restart only the channel whose settings changed
archie-core-zfb0.4. Channel settings were restart-required as a whole: a
change to one channel needed a process restart that restarted every channel.

The Messaging Service now reconciles the stored channel-settings resource on
the apply-status restamp interval (docs/prds/plugin-settings-live.md), re-layers
it over the file document exactly as boot does, and restarts only the channel
whose transport settings changed. Every other channel keeps serving.

A restart is stop-and-start, not an in-place reload: each channel gets its own
supervisor and a rebuild factory, so email and webhook -- which have no reload
seam -- are restarted too. Telegram's Start still calls dropPendingUpdates
before it long-polls, so the rule holds on the reconcile path as on /restart. A
replacement is built before the running channel is touched, so a configuration
this build cannot run leaves the running channel serving.

The composed set and the bound listen addresses are process bindings: enabling
or disabling a channel, or changing chat.webhook_addr / chat.email.listen_addr,
is refused and reported for a restart rather than applied. A channel rebuild
that fails is not recorded as applied, so it is retried on the next tick. The
applied version reaches apply status, and ChannelSettingsKind flips to live.

Files: internal/app/archiemessaging/{compose,reconcile,supervise,run}.go,
controlplane ApplyMode, archied runtime watch, configuration.md, the config
skill matrix and the regenerated dev-docs artifact.

Verified: go test ./internal/app/archiemessaging/... (race clean); go test
./cmd/archie-messaging/... ./internal/app/controlplane/... ./internal/app/archied/...
./internal/config/...; go vet and golangci-lint run on those packages (0 issues);
go fix ./... and golangci-lint fmt produced no diff; task docs:artifact:check.
2026-10-02 11:31:05 +10:00
sam cd878f3c4d fix(messaging): gate the boundary on config reads, not only package links
The deletion gate checked which packages the messaging binary links, so a
process that reads a decoded [models]/[providers] value passed while
building an LLM client in-process. Go imports are atomic, but a config
read leaves no import to measure.

- add the model-provider capability packages (transcription, embedding)
  to the banned-link table and classify links through a testable
  bannedLink rule.
- scan internal/app/archiemessaging's production sources for a read of a
  top-level config section the PRD forbids, so the projection that built
  the transcriber cannot come back.
- unit-test both rules, including a synthetic copy of the old wiring, so
  the live gate cannot rot into a no-op and still pass.

Red first: TestMessagingReadsNoBannedConfigSection failed on cfg.Models
and cfg.Providers before the capability moved, and the link gate failed on
internal/infrastructure/transcription.
2026-10-02 11:12:59 +10:00
sam b650a8c156 fix(messaging): transcribe voice notes on the model-owning side of the boundary
The Messaging Service built the voice-transcription client in-process
from [models].transcription and [providers.*], which the messaging
boundary PRD reserves to the process that owns the model runtime and its
provider credentials. The deletion gate only measured linked packages,
so the PRD violation shipped green: config reads are not imports.

Move the capability across the boundary:

- archiemessaging no longer reads [models] or [providers], builds no
  transcription client, and drops the ResolvedConfig.Transcriber field.
- telegram classifies a voice note as messaging.MediaTypeVoice and stops
  transcribing; it carries the attachment bytes and its note to the turn,
  which is all a frontend without the credential can do.
- archied builds the transcriber beside the chat runtime from its own
  model-role configuration (setupTranscriber) and gives it to every
  gateway TurnRunner.
- the turn runner transcribes a speech attachment before recording the
  inbound message, so stored history and the model context carry the
  transcript; a nil/failed/empty transcription keeps the frontend note.
- messaging.TranscribedMessageText renders the transcript into the text a
  frontend built from its media note, preserving the sender's caption.

Red first: TestTurnRunnerTranscribesSpeechOnTheModelOwningSide failed at
'Transcribe calls = 0, want 1' with the transcription call disabled.
2026-10-02 11:12:59 +10:00
sam 4241429620 feat(controlplane): add the writable review-settings resource
review.precision_gate and review.approve_before_post were file-only, so the
dashboard had no way to write them: no control-plane resource carried them.
Add the review-settings kind (docs/prds/review-settings-resource.md), seed it
from the file's [review] section, layer it field-by-field into cfg.Review,
watch it live, and publish it through ConfigView.Review. The dashboard gets a
Review settings page bound to the kind through PAGE_RESOURCES, showing each
dial beside the value in force.
2026-10-02 11:08:44 +10:00
sam cc83c9aea2 feat(agent): seed a starter SOUL and upgrade only untouched templates 2026-10-02 11:06:40 +10:00
sam 1f0cb9f318 docs(prds): name the registered step type and hold the vocabulary PRD to the provider set
docs/prds/workflow-step-vocabulary.md named the repository-rules step type
`workflow.diff-rules` while the vocabulary a deployment resolves carries
`gate.diff-rules` (workflow.DiffRulesStepName), so a definition written to
the document's word is refused at save and at run. The same decision's
sibling, docs/prds/runtime-control-plane.md, still claimed "Plugins can add
step types", which this document's closed-set rule contradicts: an
interpreted plugin contributes none, and a step type arrives as a provider
in the compiled-in set.

Correct both documents, and hold the vocabulary PRD to the provider set with
a guard test: every step-type-shaped name the document identifies must
resolve in workflowsteps.Providers(). The test fails on the pre-fix document
("workflow.diff-rules" is contributed by no provider) and passes after.

Refs: archie-core-vwqc, split from archie-core-gcoj (closed as invalidated by
re-verification: the provider set exists and cross-side agreement is already
pinned).
2026-10-02 11:05:08 +10:00
sam 19520a63e5 feat(webui): let the operator choose the retry worktree mode
A retry now carries the operator's worktree choice (refresh onto the base
branch or continue the work already pushed on the task branch) from the
dashboard through the task action to the daemon dispatch, and persists it
on the task row so it survives a reload. The daemon resolves the mode from
the task, never from the workflow; BeginRemediation records continue
explicitly so a remediation still resumes its pull request branch.
2026-10-02 10:53:21 +10:00
sam ef203ce30a fix(control-plane): unify validation for the domain-managed kinds
The control-plane catalog advertises identities, capture-events,
capture-mappings and capture-bindings as domain-managed: the domain API
validates their writes, not the generic resource store. Two of those
paths did not hold.

Identities: the file loader judged an identity name with its own
`name == ""` test, while the store derives the same string through
identity.New, whose rule rejects an empty-after-trim name. A
whitespace-only name passed the file and then failed State Store boot
(or was silently skipped by BootstrapIdentities). The loader now calls
identity.ValidDisplayName, the feature's own rule, and
BootstrapIdentities derives legacy names through
identity.FromLegacyName rather than dropping one it cannot build.

Capture mappings and bindings: mapping.Validate and binding.Validate
ran only in the archie-ui handlers, so any other client of the RPCs the
catalog names could store a mapping or binding the feature refuses.
The staterpc write handlers now apply the feature's rule before the
store, per runtime-control-plane.md ("the server validates the change,
then writes it"). The rule stays in the domain package; the handlers
only translate its refusal.

Capture-events has no file-side copy and no feature rule for a captured
event's content, so there is nothing to unify there.

Tests: identity_agreement_test.go ties the file loader's name rule to
the store constructor; domain_validation_test.go proves the mapping and
binding RPCs refuse an invalid value before it reaches the store.
2026-10-02 10:50:52 +10:00
sam ff3f80e4c1 fix(logging): file a task log record under its own event time
A remotely-shipped task log entry was re-emitted through the sink with
slog.Logger.LogAttrs, which stamps time.Now(): the entry's own event time
was discarded and replaced with the daemon's receipt time. Because a
container's records arrive in bursts, a multi-minute run collapsed onto a
few milliseconds of receipt timestamps and its timing became
unrecoverable -- which stage ran when, how long an iteration took, whether
a stage stalled.

Write now builds the record from entry.Time (falling back to receipt time
only for an entry that carries none) and dispatches it through the same
handler, preserving the sink's level gate. Receipt time stays available
separately in the record's received_at field, so event time and arrival
time cannot be confused. The agent-side producer already carried the
record's own time; a test now pins that half too.
2026-10-02 10:33:24 +10:00
sam 3da407e410 fix(archied): omit the max-token bound on reasoning-class models
The OpenAI chat-completions provider in the pinned ai-sdk serialises a
non-zero output-token bound as `max_tokens`, which reasoning-class models
reject with HTTP 400 ("Use 'max_completion_tokens' instead"). Send no bound
at all for a model the catalog marks reasoning; classic models keep their
bound. Covers the chat turn path and MCP sampling, the two call sites that
set one today.

The reasoning flag rides TurnPrepareContext from the catalog details the
gateway already resolves, so archie-core keeps one notion of "reasoning
model": messaging.ModelDetails.Reasoning. A max_completion_tokens path in
ai-sdk is still needed to ever bound these models; that is filed separately.
2026-10-02 10:25:39 +10:00
sam 4e5b91d791 fix(daemon): resume a retry whose worktree is gone
prepareWorkspace chose between Manager.Prepare (refresh onto base) and a
separate Manager.Resume, and Manager.Resume only opened an
already-prepared clone. A remediation whose local worktree the terminal
cleanup removed, an expired volume dropped, or a different host never had
could therefore never be resumed, and the retry path had no way to select
the mode that continues pushed work.

Unify the two matched methods into Prepare(..., Target): Fresh resets onto
origin/<base>, a non-empty Target resets onto origin/<that branch>, and
either target clones a missing worktree. The daemon's retry path selects
the target from the task row and passes the persisted branch, so a resume
lands on the branch the open PR lives on and fails closed when it is not
on the remote.

Delete the container-mode no-op resume (hybridTrees.Resume and the
resumableTrees capability) now that the daemon owns it. The remediate
workflow no longer stages a resume, and the retired remediate.resume step
word stays resolvable as an inert stage: a stored definition is validated
as a whole collection, so a removed word would fail every workflow to
decode, not just its own.
2026-10-02 10:24:52 +10:00
sam 12b59a046d fix(messaging): carry an inbound attachment's bytes to the gateway turn
A Telegram photo or document was dropped between the process that receives
it and the one that runs the turn. Telegram runs in archie-messaging, which
downloads the attachment; the turn runs in archie-gateway/archied, which
never holds the bot token. gatewayrpc stripped MediaAttachment.Data on the
inbound request, so by the time the Gateway's model seam ran, the
attachment had no bytes and no URL, and mediaParts dropped it -- the model
read only the "[photo]" placeholder.

- proto/gateway/v1: Media gains a `data` field, populated only on the
  inbound request. Stored-history and outbound-event directions keep it
  empty: a persisted record still strips the bytes, and a streamed media
  event still carries metadata only.
- gatewayrpc: the inbound conversion maps Data; the stored and outbound
  conversions deliberately do not.
- archied: the Gateway server raises its gRPC receive limit from the 4 MB
  default to the messaging contract's inbound attachment ceiling plus
  framing headroom, so a maximal (20 MB) attachment is accepted rather than
  failing the RPC.
- messaging: MaxInboundAttachmentBytes is the single ceiling the Telegram
  download gate and the Gateway receive limit both derive from.

Red first: a new TestInboundMediaBytesReachTheTurn asserted the bytes crossed
the wire and failed at that assertion before the mapping changed; it was then
consolidated into TestInboundMediaRoundTrip, which now pins metadata plus bytes.
2026-10-02 10:18:01 +10:00
sam af1b9855ab fix(messaging): report the daemon version this process was built from
The Telegram update-report relay verified every pending report against a
nil running-version map, so a successful update always ended with "I still
can't confirm which version is actually running" and read as a failure.

The 2026-09-19 extraction decision left RunningVersions nil because the
Messaging Service could obtain only another binary's stamp. 70ea9b2a
(archie-core-k94o) then gave every host binary, archie-messaging included,
a buildinfo.Version self-report stamped with the archied release tag, so
the process now has a truthful source for its own running version: report
daemon = buildinfo.Version and leave the agent Unverified, since every
archie-agent process is task-scoped and observed only through
daemon.AgentStatus. A relaying process that the update did not replace
still reports its old build, which Verify reads as drift.
2026-10-02 10:16:22 +10:00
sam 3e878464f9 fix(workflow): deliver the feasibility PRD to the forge issue
The deliver stage's notify() returned early when [notify].webhook was
unset, so a parked feasibility PRD left the process nowhere while both
its doc comments promised an issue comment that was never posted. A
default deployment sets no webhook: the PRD reached only the operator
who went looking.

notify now posts a one-way delivery notice on the task's forge issue
before the webhook, for a forge-backed task. The issue is the channel a
default deployment always has; the webhook remains an additional push.
Neither is a reply channel -- the decision surfaces stay chat and the
Web UI, so the notice carries the PRD and names where to decide rather
than inviting a reply, and no reply cursor is recorded (WatchCommentID
stays 0).

Both stale doc comments now describe what the code does.
2026-10-02 10:10:04 +10:00
sam 216be98652 feat(plugins): load plugin and secret-engine files without a restart
docs/prds/plugin-settings-live.md: a `plugin-settings` change now applies
live for plugin_dir, module_dir and secret_engine_dir, and a file dropped
into any of those directories loads on the reconciliation tick.

- PluginSettingsKind moves to ApplyMode "live" and joins
  runtimeResourceKinds, so a stored directory change re-layers immediately.
- A reconciler re-runs the directory loads and the layering on the
  apply-status restamp interval. Files are content-hashed, so only new and
  changed files load.
- plugin.Host grows Add and Replace for the running state (Register keeps
  its pre-Start rule), and plugin.LoadFile / secret.Registry.LoadFile are
  the single-file load paths the boot loader and the reconciler share.
- Yaegi cannot unload, so a removed file keeps running and the removal is
  reported through apply status against the live version until restart; a
  clean pass clears it.
- A plugin-settings document that changes skills_dir is refused before the
  layering, since skills_dir has no live consumer.
- applystatus.Reporter.AppliedVersion lets the reconciliation report an
  outcome without resetting the version to zero.
2026-10-02 10:01:35 +10:00
sam f77baecb00 feat(gateway): gate admin slash commands by sender policy
Add SlashAccessPolicy, a deny-by-default role policy (admin allowlist plus
per-user delegated commands), and enforce it in Router.route before any
local command dispatches. A refusal executes none of the handler, is
visible to the sender, logged, and recorded through an optional
SlashDenials sink.

No config field yet: the gatewayrpc wire drops Message.SenderID (only the
display Sender crosses), so a per-sender allowlist cannot match a Telegram
sender at the daemon Router until the layered auth mixin carries identity
(archie-core-263). Nil policy preserves today's behaviour.
2026-10-02 10:00:37 +10:00
sam 394c76b4cc feat(messaging): adapter-facing clarify, picker and approval contract
The optional-capability pattern had media and delete but no interactive
UX: a channel could ask a human to approve an action through
ApprovalRequester, but there was no domain contract for posing a
clarifying question or presenting a single-choice picker, and no
capability flag a caller could read before attempting one.

Add ClarifyRequester and PickerRequester to internal/domain/messaging
alongside ApprovalRequester, with Clarify/Picker/Approval flags on
AdapterCapabilities and ClarifierOf/PickerOf/ApproverOf lookups that
treat the capability as the contract rather than the method set, so a
mis-reporting adapter degrades instead of panicking.

TextFallback is the shared text degrade: a channel with no native
question or selection UI still carries the interaction (numbered
prompt, parsed reply) instead of failing or silently dropping it.
FormatClarifyText/FormatPickerText/ParsePickerReply hold the rendering
and matching logic once.

The approval half reuses the existing decision model: Telegram's
telegramApprover already renders a native prompt, so it now reports the
Approval capability and deliberately does not claim Clarify/Picker it
has no native UI for. Gateway re-exports the contract.

Red-first: internal/domain/messaging/interactive_test.go drives the
test double adapters (native, degrade, mis-reporting, over-claiming)
and the text rendering/parsing; telegram/approval_test.go pins the
honest capability report.
2026-10-02 09:57:44 +10:00
sam b6defc5f31 fix(workflow): park environment gate failures without dispatching a builder
The baseline stage treated every red gate as a pre-existing code defect and
dispatched a TDD repair agent. A failure the environment caused -- no Docker
daemon for the test harness, or a gate command that is not installed -- cannot
be fixed from the worktree, so the run spent a full agent budget and then
parked. A provider error during that repair run (observed: HTTP 402) replaced
the gate cause in the park reason entirely.

Classify at the producer: gateEnvironmentCause names a failure whose command
could not start (*exec.Error) or whose output names unavailable host
infrastructure (the Docker shapes from archie-core-71mq). That failure parks
with the cause and the gate output and starts no builder. A real code failure
still dispatches exactly one repair run, and every post-dispatch failure now
carries the gate output alongside the agent/provider error instead of masking
it.

Refs archie-core-rwy6.
2026-10-02 09:57:15 +10:00
sam 5e1623f82f feat(webui): the harness settings page and its setup terminal
A Settings > Runtime page lists every credential binding with the capture
state of its OAuth token set (expiry, scopes, time; never the tokens), and
offers the setup terminal for the Kit profiles the daemon advertises. The
terminal is xterm.js over the WebSocket the harness route upgrades, lazy
loaded so xterm stays out of the shell bundle. The page reports the
terminal as unavailable when the process cannot open one, rather than
offering a control that fails (archie-core-egkf.7).

Wire shapes and URL building live in ui/src/harness/harness.ts and are
covered by ui/test/harness.test.ts; ui/dist is regenerated.
2026-10-02 09:42:26 +10:00
sam 0544fa58fc feat(gateway): list the harness page in the dashboard page registry
internal/gateway/dashboard_tools.go mirrors the dashboard's route table so
the agent can navigate to every page. The new /settings/harness route needs
its entry or page_index would not offer it (archie-core-egkf.7).
2026-10-02 09:38:09 +10:00
sam 057a4a8614 feat(webui): the dashboard reads harness bindings and opens the setup terminal
The harness page lists each credential binding with its captured OAuth
state (expiry, scopes, capture time; never the tokens), and the setup
terminal route upgrades to a WebSocket and bridges it to a harness
session. The session is a narrow webui-owned HarnessTerminal seam: the
container, sandbox network and egress path belong to the daemon, so the
dashboard process cannot implement it and answers 503 until a transport
is wired (archie-core-egkf.7). Opening the terminal authorizes as the
update action on the secret, per the PRD.
2026-10-02 09:38:09 +10:00
sam 2a1ed36a26 feat(webui): a live-apply setting reads as applied, never as a restart
The per-process apply rows labelled every older applied version "Restart
pending", including a live-apply kind that needs no restart at all: the lag is
in reporting. The wording now follows the resource's apply mode. A live kind
reads "Applied" at the stored version and "Behind" on an older one, neutrally
rather than as a warning, while a restart-required kind keeps "Restart
pending".

The restart banner's condition is lifted out of the component into a plain
module so it can be tested: only a restart-required resource with a process on
an older version earns the banner, and a live-apply kind never does. The
per-process rows are wired onto the Plugins page, where a live kind's lag is
meant to show.
2026-10-02 09:37:44 +10:00
sam 605988e0a9 fix(egress): take a run's credential kind from its binding, not the Kit
The sandbox kit spec lets one credential@1 declare apiKey and oauth
together, meaning "whichever the host has bound": the published
docker/claude-code-kit does exactly that for anthropic, and codex for
openai. archie decided the kind from the Kit's declaration instead
(egress.IsOAuthManaged), so a dual-declared service always came out
OAuth-bound however the org had bound it:

  - kitrun.resolveCredentials granted an API-key binding's service the
    empty string, so the proxy had no key to inject and every request to
    a domain the Kit named carried an empty credential. The reference
    Claude Code Kit could not authenticate with an API key at all.
  - an API-key-only service whose binding named no secret resolved to ""
    through the zero SecretRef, presenting an empty credential rather than
    staying unbound.
  - the proxy fired an apiKey inject rule for an OAuth-bound service,
    putting "Bearer " on the wire, and intercepted the token endpoint of
    an API-key-bound one, whose only outcomes were a 502 on the missing
    token set, a capture the org never bound, or -- for a required
    dual-declared credential -- a refused launch when no token store was
    configured.
  - the container's SBX_CRED_*_MODE said "apikey" for an OAuth-bound
    dual-declared service, and kitrun demanded a captured OAuth token for
    an API-key-bound one whose Kit renders a credential file.

One root cause, fixed once: the org's binding decides the kind. It is
now a first-class fact (egress.CredentialKind: apikey or oauth, absent
meaning unbound), computed from the binding in resolveCredentials and
carried to every consumer of it -- the resolver grant, the container's
credential state (mode variable and credential file), and the proxy
session, whose rule sets are built from it so a rule can only fire for a
service the run carries as that same kind. A service the run does not
carry at all keeps both rule sets, so resolving it still refuses a
required credential and skips an optional one.

Declared and bound stay two independent facts, as the reported
declared-and-granted intersection requires: neither widens the other.
No secret value crosses into kit (Bound carries kinds only) and none is
logged or written to a task row.
2026-10-02 09:37:20 +10:00
sam 06e0a87a62 feat(messaging): transcribe Telegram voice notes
A Telegram voice note was stored as the placeholder "[voice message]" and
nothing turned its audio into text, so the agent could not read it. Add a
channel-neutral Transcriber capability to the messaging media layer
(internal/domain/messaging), an ai-sdk Whisper implementation behind it
(internal/infrastructure/transcription), and wire it for Telegram.

The capability is config-driven like the other model roles:
models.transcription names a provider/model, the matching [providers.*]
entry carries the credential (api_key_env or an api_key secret ref), and
a missing role, provider, class or credential degrades to today's
behaviour with a log -- never an error that fails the turn.

A transcribed note replaces its placeholder with "[voice transcription]"
followed by the transcript, preserving any caption, so stored history
shows the agent it is reading a machine transcript.
2026-10-02 09:36:48 +10:00
sam 1d2fcec69e feat(messaging): retract ephemeral replies after their TTL
An EphemeralReply is a status notice a channel delivers and then removes
once its TTL elapses, so a transient placeholder does not linger in the
chat history after it is superseded. The TTL lives on the reply, the
Delete capability joins AdapterCapabilities alongside Media, and
channels.EphemeralSender does the send-then-retract with an injected
clock. A sender that cannot delete still delivers the reply and logs the
missed retraction instead of failing the turn.

Telegram implements MessageDeleter through deleteMessage and honours the
mechanism for the "Archie reloaded." status notice; the pre-restart
"Reloading…" message remains the durable acknowledgement.
2026-10-02 09:36:16 +10:00
sam f5fbec6371 feat(container): send registry auth on image pull
ImagePull was called with empty options, so a private registry answered 401
on every pull regardless of containers.pull_policy. docker login did not help
either: it writes ~/.docker/config.json, which only the CLI reads.

Add containers.registry_auth, a SecretRef reusing the credential shape the
rest of the file uses, whose resolved value is a Docker registry.AuthConfig
JSON document. archied resolves it at boot and the pool base64url-encodes it
into the X-Registry-Auth header ImagePullOptions carries.

A missing or unresolvable credential degrades to the anonymous pull that
existed before the field (with a warning), so a deployment with no private
registry is unaffected; a half-named reference is refused by
configuration.Validate. The field is file-owned and json:"-", so the
container-runtime-policies layering has to preserve it explicitly, the way it
preserves Profiles.
2026-10-02 09:35:28 +10:00
sam a9f441042e style(mcp): apply go fix to the sampling transport 2026-10-02 09:35:21 +10:00
sam a8a04929af feat(mcp): answer server-initiated sampling from the chat model
MCP servers could not ask the client for a completion: the transports
routed only responses by id and dropped every server-initiated request,
so a server calling sampling/createMessage hung waiting for a reply.

Route server->client requests through the transport readers to a handler
on the client. The client answers sampling/createMessage from a
configured SamplingHandler and maps a missing handler, malformed params
and handler failure to JSON-RPC errors for that request alone -- never a
silent drop or a failed session. stdio and SSE implement the routing;
the stateless HTTP transport cannot receive the request at all and says
so in its doc.

The daemon wires the handler to its own chat runtime and active model
role, so a sample is answered by the model a chat turn uses, capped by
the request's maxTokens and the model's output ceiling, and recorded on
/status like every other model call. No model configured is a refused
request with a clear error.
2026-10-02 09:33:32 +10:00
sam 1e1f51af55 feat(archied): refresh the model catalog at runtime
modelcatalog.Load ran once at boot, so a provider whose credential became
resolvable later, or a model models.dev published since, needed a restart.
The daemon now re-reads the catalog on an interval and republishes everything
derived from it, without a restart.

- `refreshModelCatalog` re-reads through the same options boot used, layers
  the result with the process's own runtimeConfig, and publishes. A failed
  read keeps the catalog already running: the model list must not go blank
  because the catalog service had a bad minute.
- Boot now publishes the merged config into `config.Holder`. applyModelCatalog
  only ever wrote into the boot struct, so the catalog's model limits and
  discovered providers never reached the running config -- and therefore never
  reached `config.ForTask`, which is what a workflow stage resolves model
  limits from. The first live update was the first snapshot to carry them.
- The chat model manager's catalog is a contribution that is replaced, not
  merged. A model the catalog no longer publishes stops being offered and
  described instead of lingering from an earlier snapshot.
- The catalog snapshot and the refs it contributes live behind one lock: the
  refresh goroutine replaces them while the config paths read them.

The daemon owns the published configuration, so the loop is the daemon's; the
standalone Gateway reads the catalog once at boot, as it does for the other
live-applied settings it does not watch.

The interval is deliberately not an operator setting: the catalog is derived
state rather than a stored resource, and a knob with no second consumer would
be a setting that parses and does nothing.

A key exported into the shell after the process started stays invisible to
every refresh, and cannot be otherwise: the process environment is fixed at
exec. A credential a secret engine starts resolving is visible, because the
catalog's credential lookup is the secrets registry.

Refs archie-core-zfb0.6
2026-10-02 09:24:32 +10:00
sam 3c2b1f6654 fix(storepkg): validate every declared authority grant at the package descriptor
Descriptor.Validate duplicated only the forge-permission half of
Authority.Validate, so a descriptor could declare a blank credential
service, egress host, trigger or tool and be installed: the acceptance
record refused the same blank, leaving the descriptor the operator is
shown disagreeing with the record the package is bounded by. Delegate to
Authority.Validate, the one validation of a grant list, and cover it with
the blank cases the descriptor now has to refuse.
2026-10-02 09:20:49 +10:00
sam e10d862f6a feat(tools): checkpoint files before mutating write and edit
The built-in workspace write and edit tools replaced file content with
no recoverable pre-state. Each now snapshots its target immediately
before the write and refuses the mutation when the snapshot cannot be
taken, so a checkpoint is never silently skipped.

Checkpoints land in .git/archie-checkpoints inside the workspace, the
same per-task state directory that already holds the task brief and the
prepared sentinel: nothing under .git can be swept onto the task branch
by go-git's Add{All:true}. A checkpoint records either the file's bytes
or its absence, so creating a file restores by removing it;
RestoreLatest returns the most recent pre-state for a path.

Shell, Go-script, MCP and plugin tools are deliberately not covered:
the set of files they can touch is not enumerable from the call.
2026-10-02 09:18:47 +10:00
sam 721968eeff fix(workflow): derive a declared output's captures need for profile selection
A declared output is written through a capture tool, but the dispatch-time
harness check read only the workflow's needs.captures. A workflow that
declared outputs without also writing needs: {captures: true} could name a
Kit profile whose adapter serves no capture tools; a required output then
parked with "output %q is required" instead of the profile being refused.

WorkflowInterface.Needs now reports captures for a non-empty Outputs beside
the declared needs, and the profile check and Kit launcher read it. The
needs: block keeps its meaning for a workflow that captures without
declaring outputs.

- internal/domain/workflow/task: Needs() derives captures; DeclaredNeeds
  carries only the declared block.
- internal/daemon: pinTaskProfile and runKitTask read Needs().
- tests: parse-time derivation, and pinTaskProfile refusing an
  outputs-declaring workflow against a Kit adapter with no capture MCP.
2026-10-02 03:19:07 +10:00
sam 446d94681b feat(webui): block invalid settings and show a field's former value
The Settings save flow tracked dirty fields but did not consult validity:
each page validated in a local computed the save bar could not see, so an
invalid draft could be saved and the server refused it a round-trip later.
The rules now live once in ui/src/settings/validation.ts, the store derives
invalidFields from the same draft diff as dirtyKinds, and the save bar lists
the invalid fields and refuses Save. The route-leave guard still reads the
one dirty model, so the two can never disagree.

Every edited field also shows its former value inline ("was 1m0s") from the
same per-kind diff, via a new DraftHint component.

The per-kind diff is now a cached computed, so the hint, the save bar and
the guard share one computation per change instead of one per consumer.
2026-10-02 03:14:46 +10:00
sam 04dedde52c feat(daemon): resolve the producing identity's org before publishing work
The poller, the forge webhook intake and the review scan built envelopes
without an Org, so every task and reaction keyed on the default org no
matter which org its identity served. Resolve the org through the State
Store's PrincipalSource (the existing org RPC over the staterpc client)
on the single publish paths:

- PublishTask stamps the envelope's identity org before encoding, so the
  idempotency key is org/identity/owner/repo/number; a resolve failure
  stops the publish instead of keying under a guess.
- PublishReaction is the one reaction write path, stamping the root org
  when the webhook producer could not; the poll scan stamps its task's
  already-resolved org, so both deliveries of one review key the same.
- The webhook receiver's inline publish closure moves onto the daemon
  method, removing the second reaction write path.
2026-10-02 03:12:42 +10:00
sam a4abafe0ad docs(prds): decide the plugin live-reload open questions
Review of the plugin-settings reconcile PRD (F1-F8) returned REVISE: the doc
stated outcomes without making the decisions behind them. This revision makes
each one, so the PRD can carry the plugin live-reload work:

- ApplyMode is set to live and the gateway's version lag moves from the
  restart banner to the per-process applied-version rows.
- The skills_dir refusal gets a check site (a pre-layer comparison on the live
  path, not the value-agnostic validator), keeping #1143's validates-nothing
  intact.
- Records the four-directory coupling and why a whole-document refusal is
  acceptable.
- Justifies stop-then-start as a module's only check, and adds the refused
  dependent-replace outcome to the apply-status taxonomy.
- States that a reconcile tick re-runs layering and provider resolution, and
  what archie-gateway actually boots.
- Adds the Verification section.

Cited package paths re-checked against the tree; the control-plane
definitions now cite internal/app/controlplane/.
2026-10-02 03:10:18 +10:00
sam 6bf24d73db test(agentexec): make seeded git repos independent of host signing config
The temp-repo fixtures in agentexec, domain/workflow and agentworker seeded
a repository and committed without disabling gpg signing, so a host with
commit.gpgsign=true and an agent-only SSH signing key failed those tests at
the fixture's own git call ("Couldn't get agent socket?").

Each producer now writes commit.gpgsign=false into the temporary repository's
own config, which also covers commits made inside the seeded repo by a
subprocess (the fake harness). A regression test per package pins the
fixture's hermeticity under a hostile global git config.
2026-10-02 03:06:48 +10:00
sam 20b4595a95 style(postgres): apply the formatter to the rebased review-gate row mapping 2026-10-02 02:30:16 +10:00
sam 2341c6df36 style(taskactions): apply go fix to the gate response test
The modernizer promotes the embedded fakeStore's fields into the literal;
task fmt runs go fix, so the lane gate flags the pre-fix form.
2026-10-02 02:27:59 +10:00
sam eaabc254be feat(webui): the dashboard answers the review gate, including a re-review's instructions
The server offers re-review for a waiting task, and it refuses one
without instructions, so the control needs a dialog of its own rather
than the plain confirm: it collects the operator's focus and posts
api.taskAction(id, "rereview", {instructions}). taskAction now takes the
review-gate payload (instructions and the finding keys approve posts, an
absent selection meaning all), and the freeze-dried catalog gains
Re-review so the first render matches the served catalog. The rereview
event kinds join the live-update topic list.

ui/dist is regenerated and committed; task test:ui and the typecheck pass.
2026-10-02 02:27:59 +10:00
sam 68a17aed91 feat(taskactions): one approval path answers the review gate from every surface
Decision 1 of docs/prds/pr-review-operator-response.md: the dashboard's
POST /api/tasks/{id}/action is the authoritative surface, chat's /approve
is a shortcut that posts every offered finding, and both reach the same
Service.Apply so one operator intent cannot be recorded as two decisions.

The action body now carries instructions (required by rereview) and the
finding keys an approve posts. taskactions.Service answers an offer
through the guarded RespondReviewGate write: approve records the
selection on the offer, rereview records the instructions, increments the
round and clears the offer, and a rereview with no instructions, with no
offer, or at the cap is refused as a conflict with nothing written. Chat
reaches the same service: chatTaskControllerAdapter.ApproveChatTask
delegates to b.taskActions() instead of its own requeue, which is what
removed the handler's hardcoded implement handoff -- feasibility's deliver
stage now names implement on the task before it waits, the invariant the
handler needs.

The payload crosses every hop it must: the NATS action request, the
gateway.v1 gRPC contract (regenerated), the LocalChatAdapter, and the
webui handler, which answers 400 for a rereview without instructions and
409 for the cap. A rereview event kind names its actor the way the
approval and rejection trios do.

taskactions.Task gains review_gate and rereview_rounds, and the two
surfaces' tests assert they agree.
2026-10-02 02:27:59 +10:00
sam 4f01b1604d feat(prreview): the gate records the review it holds and posts the operator's answer
The gate now writes its offer to the task's review_gate document before it
waits -- the scored findings, the head SHA, the pull request's identity
and the resuming workflow -- because the engine cannot resume a run
partway, so a resumed pipeline recomputes its findings and only the
recorded document is the review the operator saw
(docs/prds/pr-review-operator-response.md, "The review the operator
answers").

The document lives with the task column it encodes (internal/domain/
workflow/task), the way EncodeInputs does, and carries each finding as
opaque JSON beside the key an operator's selection names. That is what
lets the dashboard render and answer a gate without linking the workflow
engine -- cmd/archie-ui/architecture_test.go measures the linked binary,
and an operator-domain service importing the finding's own type linked
prreview into archie-ui. prreview owns the key derivation and the
finding's encoding, and the MaxRereviewRounds constant lives in the task
vocabulary beside the action names it answers with.

Intake restores the document into the run: a re-review's instructions
ride one labelled block in runLens's and runReviewer's missions only, and
an approve's recorded review is posted by phase 9 filtered by the
operator's selection, with the recomputed findings kept out of the output.
The gate does not wait again on an answered approve.

taskstate gains the rereview action the vocabulary needs -- offered for
waiting_human beside approve and reject, routed to queued -- whose
surfaces follow in the response channel.
2026-10-02 02:27:59 +10:00
sam bea1c69210 feat(staterpc): carry the review gate and re-review rounds, with one guarded response write
The operator-approval gate needs its conversation with the operator
persisted before it waits and its answer threaded back
(docs/prds/pr-review-operator-response.md, Decision 2). review_gate is a
text column beside review_payload, encoded by the package that owns the
type; rereview_rounds is a bigint counted only by the new guarded write.

RespondReviewGate is the one write beside RetryTask/BeginRemediation: a
transition-table guard, one from-status check, an optional round
increment, and a requeue in one transaction. The re-review cap is
enforced in the row under the guard's lock, and a refusal leaves the
task waiting and the row untouched, surfaced as ErrRereviewCapReached
(added to the wire error table and the conformance battery). The column
joins the column-listing ListTaskSummaries projection and UpdateTask, so
the dashboard reads the offer and the store round-trips the count.
2026-10-02 02:27:59 +10:00
sam 7f51f1c39e feat(prreview): the operator-approval gate is PRReview's own stage-list entry
prReviewDecisionStages loses the withOperatorApproval flag: the shared
list the implement workflow splices (archie-core-afbk.7) never carries
the gate, and the standalone pipeline inserts it itself, immediately
before its merge gate (docs/prds/pr-review-operator-response.md, "The
gate is not reachable from archie's own PRs"). A structural test pins
both directions so a future edit cannot re-share the gate.
2026-10-02 02:27:30 +10:00
sam 3c32b63066 feat(workflow): declared workflow outputs for wait:true callers
A workflow declares named, typed outputs beside its inputs
(task.WorkflowInterface.Outputs). A run writes each one once through a
capture tool named after the output, or from a deterministic Go stage via
TaskContext.WriteOutput; the finish write persists them on tasks.outputs
before the outcome transition. A wait:true caller receives its successful
callee's outputs through WorkflowCallStatus and publishes one as its own
through a call step's outputs assignment.

Build authority: docs/prds/workflow-call-outputs.md.

- save time: WorkflowInterface.Validate checks output names and types;
  checkCallOutputs checks a call's assignment against both declarations
  and refuses outputs over a wait:false call.
- run time: undeclared, mistyped and required-but-missing values park the
  run; published callee values are re-checked against the caller's type.
- storage/wire: tasks.outputs (goose 0024), reset by each claim so the set
  is attempt-scoped; an encoded set past the structured-payload bound is
  refused, not clipped. Task.outputs_json and
  WorkflowCallStatusResponse.outputs_json carry it; task.Caller keeps its
  two methods.
- examples: firewall-contain calls firewall-investigate and publishes its
  summary as its own output.
2026-10-02 02:09:03 +10:00
325 changed files with 21837 additions and 2467 deletions

No files matched your search

@@ -120,6 +120,7 @@ Do not expose these helper flags as daemon settings.
| `models.<role>` | Workflow stages read requested role then fall back to `builder`; triage judge falls back `triage` to `planner`. | production-wired |
| `providers.<name>.class`, `api_key_env`, `base_url` | Builds ai-sdk catalogue. `base_url` rejects parse errors, userinfo, query, and fragment. | production-wired |
| `budgets.max_steps`, `max_tokens`, `wall_clock`, `gate_max_failures` | No loader defaults; `Budgets` documents zero as disabling. | production-wired |
| `review.precision_gate`, `review.approve_before_post` | 2026-10-02: one control-plane resource `review-settings` (`internal/app/controlplane/review_settings.go`), seeded from `[review]`; the stored document outranks the file, layers field-by-field into `cfg.Review` (`runtimeConfigFrom`), and is watched live (`archie/control_plane.go`). Reaches a run through `Daemon.configFor` → `ForTask` → `TaskContext.Cfg`; the dashboard writes it on the Review settings page. | production-wired |
| `diff_cap_lines` | Zero replaced by `400` in `finalize`; `StageDiffCap` treats `<= 0` as disabled. TOML zero cannot disable; negative value does. | production-wired |
| `web.listen` | Empty becomes `127.0.0.1:8484`; literal `off` disables. No authentication. | production-wired |
| `notify.webhook` | Empty disables; feasibility workflow POSTs human-decision notifications when set. | production-wired |
@@ -169,13 +170,14 @@ looking at before reporting one:
| Key | Effective behaviour |
| ----------------- | ------------------------------------------------------------------------------------------------------------------ |
| `image` | Required only when enabled. |
| `max_concurrency` | Zero means no limit; used by general task dispatcher and container pool. |
| `max_uptime` | Default `60m` only when enabled; caps container lifetime. |
| `volume_ttl` | Default `72h` only when enabled; negative rejected. |
| `pull_policy` | Default `missing` when enabled. Pool code recognises `missing` and `always`; other strings silently skip pre-pull. |
| `network` | Empty uses best-effort self-network detection. Repository overlay pins `archie-core_default`. |
| `profiles` | Named agent execution environments (image or Kit); resolved per task dispatch (`ContainerConfig.Profile`). Own control-plane resource `agent-profiles` (`json:"-"` on the field so `container-runtime-policies` never carries it), `ApplyMode: live` -- a profile change applies without a restart, unlike the rest of `[containers]`. |
| `image` | Required at boot. Read by the container pool on every acquire (`containerLiveSettings` -> `Pool.current`, archie-core-zfb0.2), so a stored change applies without a restart. |
| `max_concurrency` | Zero means no limit. Read by the pool per acquire and applied to the daemon's running task dispatcher on publish (`ResizeTaskDispatcher`). |
| `max_uptime` | Default `60m` at boot; caps container lifetime. Read by the pool per acquire, so a stored change reaps the next container under the new cap. |
| `volume_ttl` | Default `72h` only when enabled; negative rejected. Re-read per cleanup cycle. |
| `pull_policy` | Default `missing` at boot. Pool code recognises `missing` and `always`; other strings silently skip pre-pull. Read by the pool per acquire. |
| `registry_auth` | `SecretRef` (file-owned, `json:"-"`). Resolved at boot by `archied`'s `resolveRegistryAuth` and passed to `container.Config.RegistryAuth`; the pool base64url-encodes it into the `X-Registry-Auth` header (`internal/container/pool.go`'s `registryAuthHeader`). Zero ref = anonymous pull. Unresolvable/empty credential or an unencodable one warns and pulls anonymously rather than failing the pool. A half-named ref is refused by `configuration.validateContainers`. `container-runtime-policies` cannot carry a secret ref, so this stays boot-resolved and requires-restart, unlike `image`/`pull_policy`. |
| `network` | Empty uses best-effort self-network detection at boot; the repository overlay pins `archie-core_default`. Read by the pool per acquire, falling back to the boot-resolved network when the live value is empty. |
| `profiles` | Named agent execution environments (image or Kit); resolved per task dispatch (`ContainerConfig.Profile`). Own control-plane resource `agent-profiles` (`json:"-"` on the field so `container-runtime-policies` never carries it), `ApplyMode: live`. |
| `credentials` | Maps a Kit `credential@1` service to a secret (`config.CredentialBinding`). Own control-plane resource `credential-bindings`, `ApplyMode: live`, `json:"-"` on the field. `kitrun.Launcher` reads it fresh from the live `config.Holder` on every `Launch`, never a value captured at boot -- see `internal/infrastructure/kitrun/launcher.go`'s `resolveCredentials`. A binding alone grants nothing; see `org`/`granted_credentials` below. |
| `org` / `granted_credentials` (root and per-identity) | An identity's org and the `credentials` service names it may resolve. `daemon.runKitTask` reads both fresh per dispatch (`d.configFor(task)`) and passes them to `kitrun.Request`; `config.ContainerConfig.BoundCredentials` computes the declared (Kit) ∩ granted (identity) ∩ org-matched intersection -- neither declared nor granted widens the other. |
@@ -193,6 +195,13 @@ looking at before reporting one:
Telegram `/restart` re-runs same TOML overlay load, then replaces only gateway
token and allowed-user list. Every other file setting is startup-only.
A stored `channel-settings` change is reconciled live by `archie-messaging`:
on a new version it re-resolves the file, layers the store over it, and restarts
only the channel whose transport settings changed (Telegram token/allowlist,
email relay, webhook route/secret). `chat.email.listen_addr` and
`chat.webhook_addr` are bound listen addresses, so those changes, and enabling
or disabling a channel, are reported outstanding and take effect on restart.
## Treat multi-identity as partially wired
`finalize` requires each identity's `name`, `bot_user`, non-empty repos, forge
@@ -216,8 +225,9 @@ Classify multi-identity changes field by field. Never label aggregate
## Preserve worker and secret boundaries
`Config.ForTask` creates a detached root snapshot: models, budgets, dispatch,
diff cap, notification, and forge host. `TaskConfig.ToConfig` reconstructs
`Config.ForTask` creates a detached root snapshot: models, budgets, the
retry/remediation cap (`MaxRetries`), dispatch, diff cap, notification, and
forge host. `TaskConfig.ToConfig` reconstructs
only those fields. Forge tokens, provider values, NATS credential references,
repos, and infrastructure settings do not enter `TaskConfig`.
+86
View File
@@ -5,6 +5,92 @@ release's per-component sections are labelled beneath its heading.
## [Unreleased]
## [1.46.0] - 2026-10-02
### archied
- feat(workflow): declared workflow outputs for wait:true callers
- feat(prreview): the operator-approval gate is PRReview's own stage-list entry
- feat(staterpc): carry the review gate and re-review rounds, with one guarded response write
- feat(prreview): the gate records the review it holds and posts the operator's answer
- feat(taskactions): one approval path answers the review gate from every surface
- feat(webui): the dashboard answers the review gate, including a re-review's instructions
- feat(daemon): resolve the producing identity's org before publishing work
- feat(webui): block invalid settings and show a field's former value
- fix(workflow): derive a declared output's captures need for profile selection
- fix(storepkg): validate every declared authority grant at the package descriptor
- feat(archied): refresh the model catalog at runtime
- feat(mcp): answer server-initiated sampling from the chat model
- feat(container): send registry auth on image pull
- feat(messaging): retract ephemeral replies after their TTL
- feat(messaging): transcribe Telegram voice notes
- fix(egress): take a run's credential kind from its binding, not the Kit
- feat(webui): a live-apply setting reads as applied, never as a restart
- feat(webui): the dashboard reads harness bindings and opens the setup terminal
- feat(gateway): list the harness page in the dashboard page registry
- feat(webui): the harness settings page and its setup terminal
- fix(workflow): park environment gate failures without dispatching a builder
- feat(messaging): adapter-facing clarify, picker and approval contract
- feat(gateway): gate admin slash commands by sender policy
- feat(plugins): load plugin and secret-engine files without a restart
- fix(workflow): deliver the feasibility PRD to the forge issue
- fix(messaging): report the daemon version this process was built from
- fix(messaging): carry an inbound attachment's bytes to the gateway turn
- fix(daemon): resume a retry whose worktree is gone
- fix(archied): omit the max-token bound on reasoning-class models
- fix(logging): file a task log record under its own event time
- fix(control-plane): unify validation for the domain-managed kinds
- feat(webui): let the operator choose the retry worktree mode
- feat(agent): seed a starter SOUL and upgrade only untouched templates
- feat(controlplane): add the writable review-settings resource
- fix(messaging): transcribe voice notes on the model-owning side of the boundary
- fix(messaging): gate the boundary on config reads, not only package links
- feat(messaging): restart only the channel whose settings changed
- feat(container): apply container-runtime-policies live
- feat(tools): reconcile MCP servers and tool settings live
- feat(messaging): add a per-turn interactive carrier for clarify and picker
- feat(gateway): offer ask_user when the turn's channel can carry a question
- feat(telegram): carry clarify and picker through the text fallback
- fix(webui): pin the dashboard validators to the control plane's
- refactor(agent): drop the SOUL upgrade branch with no shipped template
- feat(agentworker): answer MCP sampling from the task's model
- fix(workflow): bind the remediate worktree before the builder runs
- fix(config): carry max_retries into the worker TaskConfig snapshot
- fix(agentworker): supply the guardrail engine to every task run
### archie-agent
- feat(workflow): declared workflow outputs for wait:true callers
- feat(prreview): the operator-approval gate is PRReview's own stage-list entry
- feat(staterpc): carry the review gate and re-review rounds, with one guarded response write
- feat(prreview): the gate records the review it holds and posts the operator's answer
- feat(taskactions): one approval path answers the review gate from every surface
- fix(workflow): derive a declared output's captures need for profile selection
- fix(storepkg): validate every declared authority grant at the package descriptor
- feat(mcp): answer server-initiated sampling from the chat model
- feat(container): send registry auth on image pull
- feat(messaging): retract ephemeral replies after their TTL
- feat(messaging): transcribe Telegram voice notes
- fix(egress): take a run's credential kind from its binding, not the Kit
- fix(workflow): park environment gate failures without dispatching a builder
- feat(messaging): adapter-facing clarify, picker and approval contract
- feat(plugins): load plugin and secret-engine files without a restart
- fix(workflow): deliver the feasibility PRD to the forge issue
- fix(messaging): carry an inbound attachment's bytes to the gateway turn
- fix(daemon): resume a retry whose worktree is gone
- fix(logging): file a task log record under its own event time
- fix(control-plane): unify validation for the domain-managed kinds
- feat(webui): let the operator choose the retry worktree mode
- fix(messaging): transcribe voice notes on the model-owning side of the boundary
- feat(messaging): restart only the channel whose settings changed
- feat(container): apply container-runtime-policies live
- feat(tools): reconcile MCP servers and tool settings live
- feat(messaging): add a per-turn interactive carrier for clarify and picker
- feat(agentworker): answer MCP sampling from the task's model
- fix(workflow): bind the remediate worktree before the builder runs
- fix(config): carry max_retries into the worker TaskConfig snapshot
- fix(agentworker): supply the guardrail engine to every task run
## [1.45.0] - 2026-09-30
### archied
+4 -2
View File
@@ -89,8 +89,10 @@ domain migration.
reload of `nats.url` logs an error requiring restart to prevent split-brain
delivery.
- **Container pull policy:** `containers.pull_policy` must be `"missing"`
(default) or `"always"`. Private registries return 401 on `"always"` because
no auth is sent. Refresh explicitly via `docker compose pull agent`.
(default) or `"always"`. Private registries need `containers.registry_auth`
(a `SecretRef` resolving to a `registry.AuthConfig` JSON document); without it
the pull is anonymous and a private registry answers 401. Refresh explicitly
via `docker compose pull agent`.
- **Docker Compose:** `docker-compose.yml` carries optional external NATS and
the `agent` build stanza. The agent uses a `build` profile so `up -d nats`
never starts it as a persistent service.
+44 -7
View File
@@ -57,6 +57,14 @@ var deletionGate = []bannedCategory{
{"gateway runtime", []string{
modulePath + "internal/gateway",
}},
// The PRD names "Direct LLM providers, model runtimes" among what the
// Messaging Service must not own or link. These are the in-module
// infrastructure packages that construct one; building either here means
// a model-provider credential is in the messaging process.
{"model providers", []string{
modulePath + "internal/infrastructure/transcription",
modulePath + "internal/infrastructure/embedding",
}},
{"dashboard", []string{
modulePath + "internal/webui",
}},
@@ -92,13 +100,8 @@ func TestMessagingProcessLinksNoRuntimeItOnlyDials(t *testing.T) {
if excepted(importPath, linked) {
continue
}
for _, category := range deletionGate {
for _, prefix := range category.prefixes {
if !under(importPath, prefix) {
continue
}
t.Errorf("archie-messaging links %s (%s); the Messaging Service reaches that capability over a contract, never in-process (docs/prds/messaging-service-boundary.md, deletion gate)", importPath, category.name)
}
if category, banned := bannedLink(importPath); banned {
t.Errorf("archie-messaging links %s (%s); the Messaging Service reaches that capability over a contract, never in-process (docs/prds/messaging-service-boundary.md, deletion gate)", importPath, category)
}
}
for _, exception := range gateExceptions {
@@ -113,6 +116,40 @@ func TestMessagingProcessLinksNoRuntimeItOnlyDials(t *testing.T) {
}
}
// bannedLink reports the deletion-gate category importPath falls under, or
// ("", false). Classifying a single package is the rule the unit test below
// pins, so the gate can be exercised without a live `go list`.
func bannedLink(importPath string) (category string, banned bool) {
for _, c := range deletionGate {
for _, prefix := range c.prefixes {
if under(importPath, prefix) {
return c.name, true
}
}
}
return "", false
}
// TestDeletionGateBansModelProviderCapabilities pins the PRD category that the
// voice-transcription violation slipped past: the link gate would not flag
// anything merely because the process built an LLM client, so the capability
// packages are named explicitly. gatewayrpc is checked as the negative, since
// it is the permitted contract and must not be swept up by the internal/
// infrastructure/ prefix.
func TestDeletionGateBansModelProviderCapabilities(t *testing.T) {
for _, pkg := range []string{
modulePath + "internal/infrastructure/transcription",
modulePath + "internal/infrastructure/embedding",
} {
if _, banned := bannedLink(pkg); !banned {
t.Errorf("bannedLink(%q) = not banned; the Messaging Service must not construct a model-provider capability in-process", pkg)
}
}
if category, banned := bannedLink(modulePath + "internal/infrastructure/gatewayrpc"); banned {
t.Errorf("bannedLink(gatewayrpc) = %q; the Gateway contract is permitted", category)
}
}
// excepted reports whether importPath sits under an excepted package, recording
// the ones actually linked so a stale exception fails.
func excepted(importPath string, linked map[string]bool) bool {
+433
View File
@@ -0,0 +1,433 @@
// The link gate in architecture_test.go measures which packages the Messaging
// binary links. It cannot see a process that links nothing new but still reads
// a decoded configuration value it is not allowed to own -- which is exactly
// how voice transcription was wired into the Messaging Service while a green
// link gate watched: internal/app/archiemessaging/config.go built an LLM
// transcriber from [models].transcription and [providers.*], both of which
// docs/prds/messaging-service-boundary.md reserves to the model-owning process.
//
// This file closes that shape of hole: it scans the Messaging Service's own
// production sources for a read of a top-level config section the PRD forbids,
// so re-adding the projection field fails here rather than shipping green.
package main
import (
"go/ast"
"go/parser"
"go/token"
"os"
"path/filepath"
"strings"
"testing"
)
// messagingAppDir is the Messaging Service's application package, relative to
// this package's directory (go test runs with the package dir as cwd).
const messagingAppDir = "../../internal/app/archiemessaging"
// bannedConfigFields are the top-level config.Config fields the Messaging
// Service must never read, by Go field name. The PRD's "Configuration
// Ownership" section permits [chat.*], [services.*], [secrets] and the
// path/identity keys the process needs to locate itself; every entry here is a
// capability another process owns. internal/config's field names are the
// concrete keys, because that is what a Go read names.
var bannedConfigFields = map[string]bool{
"Forge": true, // [forge]
"Repos": true, // [repos]
"Models": true, // [models]
"Providers": true, // [providers]
"Containers": true, // [containers]
"NATS": true, // [nats]
// [runners] has no top-level field in internal/config today; if one is
// added its own name must be listed here with it.
}
// configReadOffenders returns the banned "<holder>.<Field>" config reads in
// src, in source order. A holder is anything the file gives the
// internal/config.Config type: an explicitly typed parameter, var, receiver or
// struct field, a short variable declaration or inferred var whose right-hand
// side resolves to one ("cfg := doc.Config"), or a range variable over a slice
// of them. The type is tracked by name because a name is what a Go read
// selects on. A config value returned by a call this file does not declare is
// outside a syntactic rule and stays untracked.
func configReadOffenders(src []byte) []string {
fset := token.NewFileSet()
file, err := parser.ParseFile(fset, "config.go", src, parser.SkipObjectResolution)
if err != nil {
// A file that does not parse is reported as an offender rather than
// ignored: silently skipping it would make this gate blind exactly
// when the tree is broken.
return []string{"parse error: " + err.Error()}
}
kinds := configKinds(file)
var offenders []string
ast.Inspect(file, func(n ast.Node) bool {
sel, ok := n.(*ast.SelectorExpr)
if !ok || !bannedConfigFields[sel.Sel.Name] {
return true
}
switch base := sel.X.(type) {
case *ast.Ident:
if kinds[base.Name] == configValue {
offenders = append(offenders, base.Name+"."+sel.Sel.Name)
}
case *ast.SelectorExpr:
// doc.Config, or the config-typed struct field h.cfg. Config is
// named by configuration.Document in another package, so it
// cannot come out of the kinds map; treat a selected Config as
// a config holder unconditionally.
if base.Sel.Name == "Config" || kinds[base.Sel.Name] == configValue {
offenders = append(offenders, base.Sel.Name+"."+sel.Sel.Name)
}
}
return true
})
return offenders
}
// configKind is how an identifier or expression reaches a config.Config value.
type configKind int
const (
notConfig configKind = iota // not a config value
configValue // config.Config
configSlice // []config.Config, [N]config.Config or map[K]config.Config
)
// configKinds resolves the identifiers in file that hold a config.Config, in
// two passes because an inferred holder's source must be classified before the
// holder's own name can be.
func configKinds(file *ast.File) map[string]configKind {
kinds := map[string]configKind{}
ast.Inspect(file, func(n ast.Node) bool {
switch node := n.(type) {
case *ast.FuncDecl:
classifyFields(node.Recv, kinds)
classifyFields(node.Type.Params, kinds)
case *ast.ValueSpec:
classifyNames(node.Type, node.Names, kinds)
case *ast.StructType:
for _, field := range node.Fields.List {
classifyNames(field.Type, field.Names, kinds)
}
}
return true
})
// Inferred holders: a single source-order pass suffices because Go
// requires a name to be declared before it is referenced.
ast.Inspect(file, func(n ast.Node) bool {
switch node := n.(type) {
case *ast.AssignStmt:
if node.Tok != token.DEFINE {
return true
}
classifyValues(node.Lhs, node.Rhs, kinds)
case *ast.ValueSpec:
// An explicit type was handled above; here only "var cfg =
// doc.Config", which infers the same way a := does.
if node.Type != nil {
return true
}
names := make([]ast.Expr, len(node.Names))
for i, name := range node.Names {
names[i] = name
}
classifyValues(names, node.Values, kinds)
case *ast.RangeStmt:
if exprKind(node.X, kinds) != configSlice {
return true
}
if id, ok := node.Value.(*ast.Ident); ok {
kinds[id.Name] = configValue
}
}
return true
})
return kinds
}
// classifyValues records each lhs identifier that resolves to a config kind of
// the rhs expression in the same position. A shorter lhs or rhs is fine;
// position is what pairs a name with its value.
func classifyValues(lhs, rhs []ast.Expr, kinds map[string]configKind) {
for i, target := range lhs {
if i >= len(rhs) {
return
}
id, ok := target.(*ast.Ident)
if !ok {
continue
}
if kind := exprKind(rhs[i], kinds); kind != notConfig {
kinds[id.Name] = kind
}
}
}
// classifyFields records every named field in fields whose type holds a
// config.Config.
func classifyFields(fields *ast.FieldList, kinds map[string]configKind) {
if fields == nil {
return
}
for _, field := range fields.List {
classifyNames(field.Type, field.Names, kinds)
}
}
// classifyNames records names as holding the config kind of typ.
func classifyNames(typ ast.Expr, names []*ast.Ident, kinds map[string]configKind) {
kind := typeKind(typ)
if kind == notConfig {
return
}
for _, name := range names {
kinds[name.Name] = kind
}
}
// exprKind resolves the config kind of an expression from the identifiers
// already classified in kinds.
func exprKind(expr ast.Expr, kinds map[string]configKind) configKind {
switch e := expr.(type) {
case *ast.Ident:
return kinds[e.Name]
case *ast.SelectorExpr:
// doc.Config of a configuration.Document, or a struct field the file
// declared as config.Config.
if e.Sel.Name == "Config" {
return configValue
}
return kinds[e.Sel.Name]
case *ast.CompositeLit:
return typeKind(e.Type)
case *ast.StarExpr:
return exprKind(e.X, kinds)
case *ast.UnaryExpr:
return exprKind(e.X, kinds)
case *ast.ParenExpr:
return exprKind(e.X, kinds)
case *ast.IndexExpr:
if exprKind(e.X, kinds) == configSlice {
return configValue
}
}
return notConfig
}
// typeKind reports the config kind a type expression names: config.Config, or
// a slice, array or map whose element is one.
func typeKind(expr ast.Expr) configKind {
switch t := expr.(type) {
case *ast.ArrayType:
if isConfigType(t.Elt) {
return configSlice
}
case *ast.MapType:
// A map's range value is its element; classify the map itself as a
// slice-like source so a range over it yields config values.
if isConfigType(t.Value) {
return configSlice
}
case *ast.StarExpr:
return typeKind(t.X)
case *ast.ParenExpr:
return typeKind(t.X)
}
if isConfigType(expr) {
return configValue
}
return notConfig
}
// isConfigType reports whether expr names internal/config's Config type,
// directly or behind a pointer.
func isConfigType(expr ast.Expr) bool {
switch t := expr.(type) {
case *ast.SelectorExpr:
id, ok := t.X.(*ast.Ident)
return ok && id.Name == "config" && t.Sel.Name == "Config"
case *ast.StarExpr:
return isConfigType(t.X)
}
return false
}
// TestMessagingReadsNoBannedConfigSection is the live gate: every production
// source file in the Messaging Service's application package is scanned, and a
// read of a PRD-forbidden section fails the build.
func TestMessagingReadsNoBannedConfigSection(t *testing.T) {
entries, err := os.ReadDir(messagingAppDir)
if err != nil {
t.Fatalf("read %s: %v", messagingAppDir, err)
}
scanned := 0
for _, entry := range entries {
name := entry.Name()
if entry.IsDir() || !strings.HasSuffix(name, ".go") || strings.HasSuffix(name, "_test.go") {
continue
}
src, err := os.ReadFile(filepath.Join(messagingAppDir, name))
if err != nil {
t.Fatalf("read %s: %v", name, err)
}
scanned++
for _, offender := range configReadOffenders(src) {
t.Errorf("%s reads %s; the Messaging Service must not read or decode that section (docs/prds/messaging-service-boundary.md, Configuration Ownership) -- the capability belongs to the model-owning process, reached over a contract", name, offender)
}
}
// A gate that read no files would pass silently. Pin that it scanned the
// package it names.
if scanned == 0 {
t.Fatalf("no production .go files found in %s: the gate asserted nothing", messagingAppDir)
}
}
// TestConfigReadRuleCatchesTheOldWiring proves the rule fails on a synthetic
// copy of the shape that shipped: a projection built by reading
// [models].transcription and [providers.*]. Without this, the live gate's rule
// could rot into a no-op and still report a pass.
func TestConfigReadRuleCatchesTheOldWiring(t *testing.T) {
const reverted = `package archiemessaging
import (
"github.com/samcharles93/archie-core/internal/config"
"github.com/samcharles93/archie-core/internal/infrastructure/transcription"
)
func project(cfg config.Config) projection {
return projection{
models: cfg.Models,
providers: cfg.Providers,
}
}
func setupTranscriber(cfg config.Config) {
_, _ = transcription.New(cfg.Models, cfg.Providers, transcription.Options{})
}
`
got := configReadOffenders([]byte(reverted))
want := []string{"cfg.Models", "cfg.Providers", "cfg.Models", "cfg.Providers"}
if strings.Join(got, ",") != strings.Join(want, ",") {
t.Fatalf("configReadOffenders() = %v, want %v", got, want)
}
}
// TestConfigReadRuleCatchesInferredHolders covers the holder shapes the first
// cut of this rule missed. It only saw a config.Config written out as a
// parameter or var type, so a value reached through an inferred short
// variable declaration ("cfg := doc.Config") or held in a struct field
// ("h.cfg.Models") left the forbidden [models]/[providers] read intact while
// the rule saw no holder to blame. Each case here fails red against that rule:
// the live gate would have reported a pass with the violation still shipping.
func TestConfigReadRuleCatchesInferredHolders(t *testing.T) {
cases := []struct {
name string
src string
want []string
}{
{
name: "short var decl from a Document's Config field",
src: `package archiemessaging
import "github.com/samcharles93/archie-core/internal/infrastructure/configuration"
func f(doc configuration.Document) {
cfg := doc.Config
_ = cfg.Models
}
`,
want: []string{"cfg.Models"},
},
{
name: "inferred var without an explicit type",
src: `package archiemessaging
import "github.com/samcharles93/archie-core/internal/infrastructure/configuration"
func f(doc configuration.Document) {
var cfg = doc.Config
_ = cfg.Models
}
`,
want: []string{"cfg.Models"},
},
{
name: "short var decl inherits config type from another holder",
src: `package archiemessaging
import "github.com/samcharles93/archie-core/internal/config"
func f(cfg config.Config) {
other := cfg
_ = other.Providers
}
`,
want: []string{"other.Providers"},
},
{
name: "config held in a struct field",
src: `package archiemessaging
import "github.com/samcharles93/archie-core/internal/config"
type holder struct {
cfg config.Config
}
func f(h holder) {
_ = h.cfg.Providers
}
`,
want: []string{"cfg.Providers"},
},
{
name: "range variable over a slice of config",
src: `package archiemessaging
import "github.com/samcharles93/archie-core/internal/config"
func f(cfgs []config.Config) {
for _, cfg := range cfgs {
_ = cfg.Models
}
}
`,
want: []string{"cfg.Models"},
},
}
for _, tc := range cases {
t.Run(tc.name, func(t *testing.T) {
got := configReadOffenders([]byte(tc.src))
if strings.Join(got, ",") != strings.Join(tc.want, ",") {
t.Fatalf("configReadOffenders() = %v, want %v", got, tc.want)
}
})
}
}
// TestConfigReadRuleAllowsThePermittedKeys proves the rule is not a blanket
// ban on config access: the sections the PRD does permit keep passing, and a
// nested chat field is not mistaken for a top-level one.
func TestConfigReadRuleAllowsThePermittedKeys(t *testing.T) {
const permitted = `package archiemessaging
import "github.com/samcharles93/archie-core/internal/config"
func project(cfg config.Config) projection {
return projection{
telegram: cfg.Chat.Telegram,
email: cfg.Chat.Email,
gateway: cfg.Services.Get("gateway").Target,
workDir: cfg.WorkDir,
botUser: cfg.BotUser,
healthURL: cfg.Health.URL(),
}
}
`
if got := configReadOffenders([]byte(permitted)); len(got) != 0 {
t.Fatalf("configReadOffenders(permitted) = %v, want none", got)
}
}
+2 -2
View File
@@ -131,11 +131,11 @@ func (f *fakeGateway) Cancel(context.Context, string) (messaging.ChatCancellatio
func (f *fakeGateway) SetPersona(context.Context, string, string) (bool, error) { return false, nil }
func (f *fakeGateway) ApplyTaskAction(context.Context, string, int64, taskstate.Action) (messaging.TaskActionResult, error) {
func (f *fakeGateway) ApplyTaskAction(context.Context, string, int64, taskstate.Action, taskactions.ActionPayload) (messaging.TaskActionResult, error) {
return messaging.TaskActionResult{}, nil
}
func (f *fakeGateway) ApplyOperatorTaskAction(_ context.Context, actor taskactions.Actor, id int64, action taskstate.Action) (messaging.TaskActionResult, error) {
func (f *fakeGateway) ApplyOperatorTaskAction(_ context.Context, actor taskactions.Actor, id int64, action taskstate.Action, _ taskactions.ActionPayload) (messaging.TaskActionResult, error) {
f.mu.Lock()
defer f.mu.Unlock()
f.actions = append(f.actions, fmt.Sprintf("%d:%s", id, action))
+14
View File
@@ -115,6 +115,11 @@ builder = "openai/gpt-5.4"
# (curators first). Unset, or a credential that fails to resolve, degrades
# to the capability reporting unavailable -- archied still starts.
# embedding = "openai/text-embedding-3-small"
# `transcription` is optional: it turns Telegram voice notes into text before
# the turn is dispatched, behind a "[voice transcription]" marker. Read by
# archie-messaging. Unset, or a provider that does not resolve, leaves the
# voice note as its media placeholder -- the turn still runs.
# transcription = "openai/whisper-1"
[providers.openai]
class = "openai"
@@ -442,6 +447,15 @@ target = "127.0.0.1:9090" # REQUIRED. The State Store gRPC address arch
# volume_ttl = "72h" # max age for persistent repo volumes and Git clone caches
# pull_policy = "missing" # "missing" or "always"
#
# A private registry needs a credential the daemon sends on pull (an
# anonymous pull always answers 401 there, whatever the policy). The value is
# a Docker registry.AuthConfig JSON document -- {"username":"bot","password":"..."}
# ("identitytoken"/"registrytoken" are alternatives; "serveraddress" is
# optional and defaults to the image's registry). Unset means anonymous pulls,
# exactly as before this key existed; a reference that cannot be resolved
# warns and pulls anonymously rather than failing the daemon.
# registry_auth = { engine = "env", key = "ARCHIE_REGISTRY_AUTH" }
#
# Agent profiles a workflow selects with `profile: <name>`. A workflow that
# names none runs under [containers].image with every tool.
# Once the State Store holds container policies, edit profiles in the
+4 -2
View File
@@ -325,8 +325,10 @@ implemented in different packages:
synthetic issue numbers.
- Repository labels currently select workflows and mirror task state, coupling
domain transitions to forge presentation.
- Failure generally becomes `parked`, but failure classification and
retryability are not represented explicitly.
- Failure generally becomes `parked`, and the failure's classification is no
longer implicit: the persisted execution record carries the class the code
that parks recorded (the vocabulary lives in `internal/taskstate`), and
retryability is read from it rather than inferred from reason text.
- Generic observability events are emitted alongside state changes rather than
being the authoritative domain events.
+23 -7
View File
@@ -335,16 +335,32 @@ Two properties matter to anyone changing this area:
query, so boot records the last published settings and the reload re-applies
them from there.
- **Live kinds arrive on a watch.** Provider settings, model role assignments,
repository policies and scheduling policy are `ApplyMode: live`
(archie-core-zfb0.1): the daemon keeps a watch per kind and, on a new
repository policies, scheduling policy, tool settings, plugin settings,
review settings, channel settings and container runtime policies are
`ApplyMode: live`: the daemon keeps a watch per kind
(`boot.startRuntimeResourceWatches`, `runtimeResourceKinds`) and, on a new
version, re-runs `boot.runtimeConfig` over the published snapshot and
republishes through `config.Holder`. A candidate the layered document cannot
run is refused before it is published — last-known-good keeps running and
the refusal is reported through apply status. The kinds stay
restart-required while a startup-built component still holds their value
(tool, plugin, container and channel settings today); a kind joins the live
list only with a consumer that re-reads it, and the model/provider half of
it also rebuilds the gateway chat runtime's provider set and the chat model
the refusal is reported through apply status. Tool settings reconcile in this
process: the daemon diffs the stored MCP server set against the running
providers — connecting added servers, disconnecting removed ones and
reconnecting changed ones while an unchanged server keeps running — and
rebuilds the `web_fetch` and `minimax` entries, whose config was captured at
construction (`boot.reconcileToolSettings`). Container runtime policies
join the live set because their consumers re-read the published config: the
container pool reads image, pull policy, network, max uptime and concurrency
cap on every acquire and the daemon resizes its running task dispatcher from
the same publish; the fields the document cannot carry, `registry_auth`
above all, stay boot-resolved and remain requires-restart. A kind whose
consumer lives in another process reconciles there too: `archie-messaging`
re-reads the stored channel settings on the apply-status restamp interval
and restarts only the channel whose transport settings changed, while the
listen addresses it has bound and the set of channels it composed stay
process bindings that take effect on restart. A kind joins the live list
only with a consumer that re-reads it; one whose startup-built consumer
never does stays restart-required, and the model/provider half of the live
set also rebuilds the gateway chat runtime's provider set and the chat model
list the chat surfaces offer.
## Completion criteria
+19 -3
View File
@@ -202,13 +202,13 @@ actually owned, not by where it used to be built:
| Seam | Resolution |
| ---------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `Version` | Sourced from the Gateway via `ChatSnapshot.Version`. The Messaging Service never stamps or infers a build version; `cmd/archie-messaging` deliberately takes no version ldflags, so it cannot report a partial upgrade as a matched one. |
| `Version` | Sourced from the Gateway via `ChatSnapshot.Version`. The Messaging Service never renders its own build into `/version`; `cmd/archie-messaging` does carry a `buildinfo` release stamp (added by `archie-core-k94o`, see the amendment below) for the updater's skew check and update verification, but `/version` still shows the Gateway's block so a partially upgraded deployment cannot report itself as matched. |
| `Updates` (`/update`) | Built in the Messaging Service from `[chat.telegram].update_check_command` / `update_install_command` — its own configuration. `Enrich` is dropped: `componentInstallTypeEnricher` reads `[nats]`, which this service must not decode, so the check command's own reported install type stands unmodified. |
| `UpdateReportPath` | A local state file under `work_dir`, keyed by the same `sha256(bot_user)[:8]` scheme the daemon used, so a pending report is found rather than written afresh. |
| `ReleaseAnnouncements` | **Deliberately left nil**, for the same reason as `RunningVersions`. `releaseannounce.Announcer` announces nothing unless each `Component` carries a parseable version, and the only versions this service can obtain are the Gateway's own build stamps, not archied's. Wiring the announcer without them yields a no-op that reads as configured, so it stays unwired until component self-reporting exists. |
| `ReleaseAnnouncements` | **Still left nil**, but no longer for the `RunningVersions` reason (that seam now self-reports — see the amendment below). `releaseannounce.Announcer` announces nothing unless each `Component` carries a parseable version, and the only version this process can vouch for is its own daemon-release stamp, not the agent's. Wiring the announcer without them yields a no-op that reads as configured, so it stays unwired until component self-reporting exists. |
| `SetShowToolCalls` | Projected from `[chat].show_tool_calls`. |
| `Reload` | Local: re-resolves the token and allowlist from this service's own config file and overlay. Reload was never a daemon fact. |
| `RunningVersions` | **Deliberately left nil.** It exists to turn an installer's claim into a checked one, and only a component's own compiled-in build can vouch for it. The Messaging Service knows neither archied's build nor the observed archie-agent version (`daemon.AgentStatus`), and the Gateway's own `gatewayVersion` is a different binary's stamp — reporting it as the daemon's would manufacture exactly the false success the check exists to catch. Update reports therefore relay as unverified claims until a component self-report contract exists. |
| `RunningVersions` | **Wired from this process's own release stamp** (`archie-core-ot33`; see the amendment below). `archie-messaging` is one of the processes in the archied release set (the installer's `GATEWAY_SERVICES`), built from the same `archied/v*` tag and stamped with `buildinfo.Version` by every build site, so its compiled-in build is evidence that the release the installer claims actually reached this host. It reports `daemon = buildinfo.Version` and reports nothing for `agent` (task-scoped, observed only through `daemon.AgentStatus`), leaving the agent Unverified. The Gateway's rendered block is deliberately not used: that is a different process's stamp. |
| `Dangerous` | Stays nil. No daemon composition ever set it; `/rollback` and `/stop` reported "not configured" before the extraction and still do. |
The Messaging Service consequently reads `work_dir`, `bot_user`, `[chat]` and
@@ -217,6 +217,22 @@ The Messaging Service consequently reads `work_dir`, `bot_user`, `[chat]` and
polls from outside to confirm an update came back up; the rest of the PRD's
prohibition list is untouched.
**Amended 2026-10-02 (`archie-core-ot33`, Telegram `RunningVersions`).** The
2026-09-19 decision left `RunningVersions` nil because it judged this process
could obtain only another binary's stamp. That was true when written and stopped
being true three days later: `ec0f17a7` (`archie-core-k94o`) gave every host
binary, `archie-messaging` included, a `buildinfo.Version` self-report so the
updater could detect per-binary skew. `cmd/archie-messaging` now registers
`buildinfo`'s `-version`, and every build site (`install.sh`,
`scripts/archie-update-install`, `.github/workflows/deploy.yml`, `Taskfile.yml`)
stamps it with the same `archied/v*` release as the rest of `GATEWAY_SERVICES`.
The messaging process therefore vouches for the `daemon` component from
`buildinfo.Version` — its own compiled-in build, never the Gateway's rendered
block — and reports nothing for the agent. A relayed update report can now
confirm the daemon and still names the agent as unchecked, while a binary the
update did not replace reads as drift. `ReleaseAnnouncements` stays nil: it needs
a version for every component, and this process can vouch only for itself.
**Phase 4 messaging severance: what must move, and the decision the PRD does
not cover (2026-09-22, `archie-core-1ng1`).** The PRD's deletion gate cannot
pass today. `cmd/archie-messaging/architecture_test.go` does not exist, and the
+2 -2
View File
@@ -314,7 +314,7 @@
"properties": {
"data": {
"contentEncoding": "base64",
"description": "Data carries the attachment bytes in-process. An inbound path sets\nit after downloading the file; outbound senders never set it and\nread Path or URL instead. It never crosses a wire or persistence\nboundary -- it is a turn-scoped value, not a stored one.",
"description": "Data carries the attachment bytes in-process. An inbound path sets\nit after downloading the file; outbound senders never set it and\nread Path or URL instead. It crosses the wire in one direction only:\nthe inbound request from a channel frontend to the Gateway process\nthat runs the turn, because that process never holds the platform\ncredential. It is never persisted, and the stored-history and\noutbound-event directions strip it.",
"type": "string"
},
"duration": {
@@ -350,7 +350,7 @@
"type": "object"
},
"type": {
"description": "Type is \"image\", \"video\", \"audio\", or \"document\".",
"description": "Type is \"image\", \"video\", \"audio\", \"voice\", or \"document\".",
"type": "string"
},
"url": {
+11 -11
View File
@@ -16,7 +16,7 @@
"url": "/dev/architecture/agent-system/",
"title": "Agent System",
"section": "Architecture",
"body": "# Agent System\n\n**Status:** Current-state investigation in progress \n**Date:** 2026-07-28 \n**Tracking issue:** [#73](https://github.com/samcharles93/archie-core/issues/73)\n\n## Domain framing\n\nThe domain is the Agent System.\n\n`Task` is not a core Agent System concept. Forge tasks, forge issues, Jira\nissues, email, Telegram messages, webhooks, and future external inputs are\nchannel-specific messages. Channels translate those messages into requests to\nthe Agent System; external issue or task models MUST NOT shape execution state.\n\nThe current `store.Task` is a persisted workflow execution. Treating that row as\na task-domain aggregate would incorrectly turn Archie into an agentic issue\nboard.\n\n## Canonical agent interaction model\n\nAn **Agent** is a persistent assistant selected for an interaction. It is not a\nmodel invocation and is not a WorkflowExecution.\n\nEach Agent has its own:\n\n- immutable acting `IdentityID` for ownership and audit;\n- durable memory and learned information about the user;\n- perspective, specialisation, instructions, and communication behaviour;\n- conversation history and channel bindings;\n- available tools, capabilities, and policy;\n- continuity across model or provider changes.\n\nThe underlying model is a replaceable runtime dependency. Changing a model,\nprovider, or model version MUST NOT create a new Agent, discard its memory, or\nchange ownership of its history.\n\nA user MAY create multiple Agents. Each Agent develops its own memory and view\nof the user from the interactions and responsibilities assigned to it. Memories\nMUST NOT be silently pooled merely because Agents belong to the same user or\napplication.\n\nAgent and Identity are separate aggregates. Every Agent references exactly one\nacting IdentityID, and no two Agents share an agent identity. User, system, and\nservice-account identities may exist without an Agent.\n\nAn action initiated through an Agent records the Agent's acting IdentityID and,\nwhen applicable, the initiating user's IdentityID. A future fully automated\nWorkflowExecution MAY use an explicitly selected service-account identity for a\nconcrete unattended use case; this is not the default execution identity.\n\n## Workflow execution identity\n\nA Workflow is owned by an IdentityID. By default, every WorkflowExecution acts\nas the Workflow owner's identity.\n\nFuture delegated execution MAY add an explicit `RunAsIdentityID` assignment to a\nWorkflow. This allows a Workflow owner to select another user or service-account\nidentity for execution without changing Workflow ownership. Until that\ncapability is deliberately introduced, owner identity and execution identity are\nthe same.\n\nWorkflowExecution audit records MUST preserve the Workflow owner, effective\nexecution identity, initiating user when one exists, and any explicit delegation\nthat selected a different run-as identity.\n\n## Memory scopes\n\nThe Agent System requires four distinct durable memory scopes:\n\n- **global shared memory**: information intentionally available across the\n Archie application;\n- **agent-wide memory**: general expertise, operating knowledge, and perspective\n owned by one Agent and safe for that Agent to use across users;\n- **user-wide memory**: information associated with one user and available to\n authorised Agents serving that user;\n- **Agent-user relationship memory**: private history and learned context from\n one particular Agent's relationship with one particular user.\n\nAgent-wide memory is private to its owning Agent by default. Another Agent MUST\nNOT read, inherit, search, or mutate it implicitly.\n\nAgent-user relationship memory is addressed by both AgentID and User IdentityID.\nNeither another Agent serving the same user nor another user speaking to the\nsame Agent may access it implicitly.\n\nArchie is a multi-user system even when initially deployed for one person.\nMultiple users, including family members, may interact with the same configured\nAgents. Per-user memory is therefore an isolation and personalisation boundary,\nnot merely an attribution tag on the primary user's records.\n\nAn Agent MUST resolve the initiating user before assembling memory context.\nMemory belonging to one user MUST NOT become visible to another user merely\nbecause both users interact with the same Agent.\n\nShared memory does not erase provenance. Every memory record retains its scope,\nauthor or acting identity, originating user when applicable, source, and\nrevision history. The rules for selecting a write scope, granting access\nto global and user-wide memory, and authorising explicit cross-scope sharing are\nsettled by [`docs/prds/memory-engine-unification.md`](/dev/prds/memory-engine-unification/)\nand recorded in [`migration-decisions.md`](/dev/architecture/migration-decisions/#5-memory-placement-and-storage)\nsection 5.\n\nAn Agent interaction may answer conversationally, use a tool or capability, read\norLine truncated
"body": "# Agent System\n\n**Status:** Current-state investigation in progress \n**Date:** 2026-07-28 \n**Tracking issue:** [#73](https://github.com/samcharles93/archie-core/issues/73)\n\n## Domain framing\n\nThe domain is the Agent System.\n\n`Task` is not a core Agent System concept. Forge tasks, forge issues, Jira\nissues, email, Telegram messages, webhooks, and future external inputs are\nchannel-specific messages. Channels translate those messages into requests to\nthe Agent System; external issue or task models MUST NOT shape execution state.\n\nThe current `store.Task` is a persisted workflow execution. Treating that row as\na task-domain aggregate would incorrectly turn Archie into an agentic issue\nboard.\n\n## Canonical agent interaction model\n\nAn **Agent** is a persistent assistant selected for an interaction. It is not a\nmodel invocation and is not a WorkflowExecution.\n\nEach Agent has its own:\n\n- immutable acting `IdentityID` for ownership and audit;\n- durable memory and learned information about the user;\n- perspective, specialisation, instructions, and communication behaviour;\n- conversation history and channel bindings;\n- available tools, capabilities, and policy;\n- continuity across model or provider changes.\n\nThe underlying model is a replaceable runtime dependency. Changing a model,\nprovider, or model version MUST NOT create a new Agent, discard its memory, or\nchange ownership of its history.\n\nA user MAY create multiple Agents. Each Agent develops its own memory and view\nof the user from the interactions and responsibilities assigned to it. Memories\nMUST NOT be silently pooled merely because Agents belong to the same user or\napplication.\n\nAgent and Identity are separate aggregates. Every Agent references exactly one\nacting IdentityID, and no two Agents share an agent identity. User, system, and\nservice-account identities may exist without an Agent.\n\nAn action initiated through an Agent records the Agent's acting IdentityID and,\nwhen applicable, the initiating user's IdentityID. A future fully automated\nWorkflowExecution MAY use an explicitly selected service-account identity for a\nconcrete unattended use case; this is not the default execution identity.\n\n## Workflow execution identity\n\nA Workflow is owned by an IdentityID. By default, every WorkflowExecution acts\nas the Workflow owner's identity.\n\nFuture delegated execution MAY add an explicit `RunAsIdentityID` assignment to a\nWorkflow. This allows a Workflow owner to select another user or service-account\nidentity for execution without changing Workflow ownership. Until that\ncapability is deliberately introduced, owner identity and execution identity are\nthe same.\n\nWorkflowExecution audit records MUST preserve the Workflow owner, effective\nexecution identity, initiating user when one exists, and any explicit delegation\nthat selected a different run-as identity.\n\n## Memory scopes\n\nThe Agent System requires four distinct durable memory scopes:\n\n- **global shared memory**: information intentionally available across the\n Archie application;\n- **agent-wide memory**: general expertise, operating knowledge, and perspective\n owned by one Agent and safe for that Agent to use across users;\n- **user-wide memory**: information associated with one user and available to\n authorised Agents serving that user;\n- **Agent-user relationship memory**: private history and learned context from\n one particular Agent's relationship with one particular user.\n\nAgent-wide memory is private to its owning Agent by default. Another Agent MUST\nNOT read, inherit, search, or mutate it implicitly.\n\nAgent-user relationship memory is addressed by both AgentID and User IdentityID.\nNeither another Agent serving the same user nor another user speaking to the\nsame Agent may access it implicitly.\n\nArchie is a multi-user system even when initially deployed for one person.\nMultiple users, including family members, may interact with the same configured\nAgents. Per-user memory is therefore an isolation and personalisation boundary,\nnot merely an attribution tag on the primary user's records.\n\nAn Agent MUST resolve the initiating user before assembling memory context.\nMemory belonging to one user MUST NOT become visible to another user merely\nbecause both users interact with the same Agent.\n\nShared memory does not erase provenance. Every memory record retains its scope,\nauthor or acting identity, originating user when applicable, source, and\nrevision history. The rules for selecting a write scope, granting access\nto global and user-wide memory, and authorising explicit cross-scope sharing are\nsettled by [`docs/prds/memory-engine-unification.md`](/dev/prds/memory-engine-unification/)\nand recorded in [`migration-decisions.md`](/dev/architecture/migration-decisions/#5-memory-placement-and-storage)\nsection 5.\n\nAn Agent interaction may answer conversationally, use a tool or capability, read\norLine truncated
},
{
"url": "/dev/architecture/bindings/",
@@ -28,7 +28,7 @@
"url": "/dev/architecture/configuration/",
"title": "Configuration",
"section": "Architecture",
"body": "# Configuration\n\n**Status:** Approved; existing-type assignment is ongoing \n**Date:** 2026-07-28 \n**Tracking issue:** [#73](https://github.com/samcharles93/archie-core/issues/73)\n\n## Ownership\n\nConfiguration has two separate responsibilities:\n\n1. `internal/infrastructure/configuration` loads files, environment variables,\n secret references, overlays, and compatibility representations.\n2. Each domain and plugin owns the typed runtime settings required for its\n behaviour.\n\n`internal/app` translates external input into those settings, validates\ncross-capability combinations, and supplies each component only its own\nsettings.\n\nThe following are prohibited:\n\n- passing complete application configuration into a domain;\n- making a domain depend on TOML, YAML, environment-variable names, secret\n references, or a configuration library;\n- defining a central runtime configuration type imported across the system;\n- placing plugin options in an unrelated global configuration model.\n\n## Private input document\n\n`internal/infrastructure/configuration` MAY use a complete decoded document\nwhile discovering files, decoding sections, and applying overlays. It is an\ninput-boundary DTO only.\n\nThe document:\n\n- remains private to the configuration infrastructure;\n- is composed from separately defined feature sections;\n- is immediately translated by application composition;\n- is never returned as the application's runtime model;\n- is never passed through runtime components.\n\nExternal DTOs and domain runtime settings are separate types.\n\nPlugins supply external configuration definitions through a narrow extension\ncontract. The core loader MUST NOT gain a field or recognised filename for every\nplugin.\n\nInvalid configuration input MUST NOT replace active valid settings or terminate\nunaffected capabilities. Configuration changes use the candidate validation,\npromotion, audit, and recovery requirements in\n[safe-change-and-recovery.md](/dev/architecture/safe-change-and-recovery/).\n\n## Dissolution of `internal/config`\n\n`internal/config` has no target location. It MUST be dissolved and MUST NOT\nsurvive as a shared package, configuration domain, or renamed global model.\n\nIts loading behaviour moves to `internal/infrastructure/configuration`. Existing\ntypes and methods are reassigned by behaviour. Repository execution policy,\ndispatch policy, retry policy, ecosystem defaults, tool policy, memory settings,\nprovider settings, and channel settings move to their\nrespective owners.\n\n## Repository settings\n\nThe existing `config.Repo` MUST NOT move intact into a repository domain or\nshared settings package. A managed repository is an application association:\n\n- work intake owns which repository references an identity monitors;\n- task execution owns gates, preflight, test protection, ecosystem behaviour,\n and change limits;\n- task lifecycle owns retry behaviour;\n- scheduling owns concurrency policy;\n- workspace management owns base branch and persistent-storage requirements.\n\nApplication composition correlates these settings by repository reference. No\ndomain receives a universal repository profile.\n\n## Documentation\n\nGenerated configuration reference describes external sections and their owning\ncapabilities. It MUST NOT establish a global Go configuration structure as the\npublic contract.\n\n## Value classification rule\n\nEvery value that controls behaviour MUST have an explicit owner and MUST be\nclassified as exactly one of:\n\n1. **Invariant** — changing it changes the meaning of the program, a wire\n protocol, or a domain concept. It may be a Go constant.\n2. **Runtime setting** — an operator, identity, domain, plugin, or deployment\n may legitimately choose a different value. It is typed configuration owned by\n the affected capability.\n3. **Policy value** — it expresses what is allowed, limited, routed, retried,\n protected, or gated. Its typed definition belongs to the owning domain and it\n is evaluated through the policy system.\n4. **Derived value** — it is computed from other settings and is not\n independently mutable.\n5. **Runtime state** — it changes as the application operates, such as the\n active model for a conversation. It is changed by a domain command and is not\n configuration merely because it can be changed.\n\nAn unclassified behavioural literal is a defect.\n\n`const` MUST NOT be used merely because a value currently has one default.\nConstants are reserved for actual invariants: enum discriminants, protocol\nversions, fixed compatibility tokens, and values whose alteration would change\nthe program's semantics rather than configure an installation.\n\nIn Go, a constant is a compile-time value and does not represent mutable runtime\nstorage. It generally has no address that can be observed or changed.\nArchitecturally, a constant therefore asserts that the value cannot change for\nthe lifetiLine truncated
"body": "# Configuration\n\n**Status:** Approved; existing-type assignment is ongoing \n**Date:** 2026-07-28 \n**Tracking issue:** [#73](https://github.com/samcharles93/archie-core/issues/73)\n\n## Ownership\n\nConfiguration has two separate responsibilities:\n\n1. `internal/infrastructure/configuration` loads files, environment variables,\n secret references, overlays, and compatibility representations.\n2. Each domain and plugin owns the typed runtime settings required for its\n behaviour.\n\n`internal/app` translates external input into those settings, validates\ncross-capability combinations, and supplies each component only its own\nsettings.\n\nThe following are prohibited:\n\n- passing complete application configuration into a domain;\n- making a domain depend on TOML, YAML, environment-variable names, secret\n references, or a configuration library;\n- defining a central runtime configuration type imported across the system;\n- placing plugin options in an unrelated global configuration model.\n\n## Private input document\n\n`internal/infrastructure/configuration` MAY use a complete decoded document\nwhile discovering files, decoding sections, and applying overlays. It is an\ninput-boundary DTO only.\n\nThe document:\n\n- remains private to the configuration infrastructure;\n- is composed from separately defined feature sections;\n- is immediately translated by application composition;\n- is never returned as the application's runtime model;\n- is never passed through runtime components.\n\nExternal DTOs and domain runtime settings are separate types.\n\nPlugins supply external configuration definitions through a narrow extension\ncontract. The core loader MUST NOT gain a field or recognised filename for every\nplugin.\n\nInvalid configuration input MUST NOT replace active valid settings or terminate\nunaffected capabilities. Configuration changes use the candidate validation,\npromotion, audit, and recovery requirements in\n[safe-change-and-recovery.md](/dev/architecture/safe-change-and-recovery/).\n\n## Dissolution of `internal/config`\n\n`internal/config` has no target location. It MUST be dissolved and MUST NOT\nsurvive as a shared package, configuration domain, or renamed global model.\n\nIts loading behaviour moves to `internal/infrastructure/configuration`. Existing\ntypes and methods are reassigned by behaviour. Repository execution policy,\ndispatch policy, retry policy, ecosystem defaults, tool policy, memory settings,\nprovider settings, and channel settings move to their\nrespective owners.\n\n## Repository settings\n\nThe existing `config.Repo` MUST NOT move intact into a repository domain or\nshared settings package. A managed repository is an application association:\n\n- work intake owns which repository references an identity monitors;\n- task execution owns gates, preflight, test protection, ecosystem behaviour,\n and change limits;\n- task lifecycle owns retry behaviour;\n- scheduling owns concurrency policy;\n- workspace management owns base branch and persistent-storage requirements.\n\nApplication composition correlates these settings by repository reference. No\ndomain receives a universal repository profile.\n\n## Documentation\n\nGenerated configuration reference describes external sections and their owning\ncapabilities. It MUST NOT establish a global Go configuration structure as the\npublic contract.\n\n## Value classification rule\n\nEvery value that controls behaviour MUST have an explicit owner and MUST be\nclassified as exactly one of:\n\n1. **Invariant** — changing it changes the meaning of the program, a wire\n protocol, or a domain concept. It may be a Go constant.\n2. **Runtime setting** — an operator, identity, domain, plugin, or deployment\n may legitimately choose a different value. It is typed configuration owned by\n the affected capability.\n3. **Policy value** — it expresses what is allowed, limited, routed, retried,\n protected, or gated. Its typed definition belongs to the owning domain and it\n is evaluated through the policy system.\n4. **Derived value** — it is computed from other settings and is not\n independently mutable.\n5. **Runtime state** — it changes as the application operates, such as the\n active model for a conversation. It is changed by a domain command and is not\n configuration merely because it can be changed.\n\nAn unclassified behavioural literal is a defect.\n\n`const` MUST NOT be used merely because a value currently has one default.\nConstants are reserved for actual invariants: enum discriminants, protocol\nversions, fixed compatibility tokens, and values whose alteration would change\nthe program's semantics rather than configure an installation.\n\nIn Go, a constant is a compile-time value and does not represent mutable runtime\nstorage. It generally has no address that can be observed or changed.\nArchitecturally, a constant therefore asserts that the value cannot change for\nthe lifetiLine truncated
},
{
"url": "/dev/architecture/dashboard-agent-tools/",
@@ -64,7 +64,7 @@
"url": "/dev/architecture/migration-decisions/",
"title": "Architecture Migration Decisions",
"section": "Architecture",
"body": "# Architecture Migration Decisions\n\n**Status:** Active migration inventory \n**Date:** 2026-07-28 \n**Tracking issue:** [#73](https://github.com/samcharles93/archie-core/issues/73)\n\n## Purpose\n\nThis document records what remains to be decided before Archie's current\nimplementation can be migrated into the approved domain-oriented structure.\n\nThe product and core domain model are not being redesigned. Existing Archie\nbehaviour and the existing session implementation are the migration\nbaseline. Remaining decisions MUST be grounded in current packages, consumers,\ntests, persistence, and runtime wiring.\n\n## Confirmed domain packages\n\n```text\ninternal/domain/\n identity/ (approved, not yet materialised as a package)\n agent/ (approved, not yet materialised as a package)\n workflow/\n messaging/ (canonical Message/Conversation types only; gateway migration pending, see section 2)\n workintake/\n plugin/ (approved, not yet materialised as a package)\n curator/\n mapping/\n scheduling/\n```\n\n- `identity` owns persistent actors, lifecycle, ownership, attribution, and\n audit identity.\n- `agent` owns persistent assistants, specialisation, instructions,\n model-independent continuity, capability context, and associations with\n identity and memory.\n- `workflow` is a separate domain. It owns reusable, versioned Workflow\n definitions, WorkflowSteps, WorkflowExecutions, StepExecutions, Attempts,\n outcomes, and typed Workflow plugin contracts.\n- `messaging` owns canonical Messages, Conversations, branches, channel and\n source correlation, replies, acknowledgements, delivery attempts, events, and\n projections.\n- `workintake` owns admission, validation, deterministic routing, and accepted\n work requests for the optional transition into durable workflow-backed\n execution.\n- `plugin` owns generic plugin identity, discovery, compatibility, and lifecycle\n mechanics. It remains metadata-only; capability semantics and typed\n registration belong to the owning domain.\n- `curator` owns the curator engine family: long-running agent loops that\n reason over memory and perform maintenance tasks on a declared shape\n (check-in interval, tool set, attached memory engine). See epic\n archie-core-yp9.\n- `mapping` owns the payload field-mapping vocabulary: binding named fields to\n JSON paths inside a captured webhook payload, and detecting drift. See\n `docs/prds/payload-field-mapping.md`.\n- `scheduling` owns the ticker engine: deciding when a scheduled job runs and\n with what concurrency. It declares `JobSource`/`Runner` contracts and owns\n no storage or delivery. See epic archie-core-1786637496320-249-be6d9a20.\n\n**Reconciliation (2026-08-31):** `curator`, `mapping`, and `scheduling` existed\nin the tree without being added to this list when their epics landed. They are\nconfirmed as legitimate, narrowly-scoped domains under the same rules as the\noriginal six — each owns one capability family, declares typed contracts, and\nhas no cross-domain reach. This list was out of date, not the tree.\n\nConversely, `identity`, `agent`, `messaging`, and `plugin` are approved target\ndomains that do not yet exist as packages. Their behaviour currently lives in\npre-migration packages (see \"Complete package destination map\" below for the\nfull inventory). Do **not** create empty stub packages for them ahead of an\nactual migration PR — an unused stub is dead code and gives no test coverage.\nEach materialises when its corresponding migration decision below is resolved\nand implemented in the same change.\n\nMemory's final package placement remains undecided pending focused review of the\nexisting implementation.\n\nThe following are approved shared or composition packages, not domains:\n\n```text\ninternal/eventbus/\ninternal/policy/\ninternal/infrastructure/\ninternal/app/archied/\ninternal/app/agentworker/\n```\n\n## Remaining migration decisions\n\n### 1. Complete package destination map\n\nEvery current package MUST receive one explicit disposition:\n\n- move into a confirmed domain;\n- move under infrastructure as an implementation of a domain-owned contract;\n- move into application composition;\n- merge with an identified duplicate;\n- remain as an approved shared application contract;\n- move to a plugin, extra, deployment, example, or generated artifact;\n- be deleted after its replacement is proven.\n\nThe largest unreviewed areas include memory, tools and capabilities, workspace\nand worktree behaviour, gates, storage, events, web UI, containers, RPC\npackages, release management, and scheduling.\n\nThe destination matrix MUST name the behaviour owner, target path, dependencies\nto remove, state and contracts to preserve, migration prerequisites, and\ndeletion criteria.\n\n**`internal/webui` / UI Service — RATIFIED Phase 3 disposition (2026-09-07).**\nThe current owner is the in-process HTTP/SSE dashboard: SPA hosting, operator\nauthLine truncated
"body": "# Architecture Migration Decisions\n\n**Status:** Active migration inventory \n**Date:** 2026-07-28 \n**Tracking issue:** [#73](https://github.com/samcharles93/archie-core/issues/73)\n\n## Purpose\n\nThis document records what remains to be decided before Archie's current\nimplementation can be migrated into the approved domain-oriented structure.\n\nThe product and core domain model are not being redesigned. Existing Archie\nbehaviour and the existing session implementation are the migration\nbaseline. Remaining decisions MUST be grounded in current packages, consumers,\ntests, persistence, and runtime wiring.\n\n## Confirmed domain packages\n\n```text\ninternal/domain/\n identity/ (approved, not yet materialised as a package)\n agent/ (approved, not yet materialised as a package)\n workflow/\n messaging/ (canonical Message/Conversation types only; gateway migration pending, see section 2)\n workintake/\n plugin/ (approved, not yet materialised as a package)\n curator/\n mapping/\n scheduling/\n```\n\n- `identity` owns persistent actors, lifecycle, ownership, attribution, and\n audit identity.\n- `agent` owns persistent assistants, specialisation, instructions,\n model-independent continuity, capability context, and associations with\n identity and memory.\n- `workflow` is a separate domain. It owns reusable, versioned Workflow\n definitions, WorkflowSteps, WorkflowExecutions, StepExecutions, Attempts,\n outcomes, and typed Workflow plugin contracts.\n- `messaging` owns canonical Messages, Conversations, branches, channel and\n source correlation, replies, acknowledgements, delivery attempts, events, and\n projections.\n- `workintake` owns admission, validation, deterministic routing, and accepted\n work requests for the optional transition into durable workflow-backed\n execution.\n- `plugin` owns generic plugin identity, discovery, compatibility, and lifecycle\n mechanics. It remains metadata-only; capability semantics and typed\n registration belong to the owning domain.\n- `curator` owns the curator engine family: long-running agent loops that\n reason over memory and perform maintenance tasks on a declared shape\n (check-in interval, tool set, attached memory engine). See epic\n archie-core-yp9.\n- `mapping` owns the payload field-mapping vocabulary: binding named fields to\n JSON paths inside a captured webhook payload, and detecting drift. See\n `docs/prds/payload-field-mapping.md`.\n- `scheduling` owns the ticker engine: deciding when a scheduled job runs and\n with what concurrency. It declares `JobSource`/`Runner` contracts and owns\n no storage or delivery. See epic archie-core-1786637496320-249-be6d9a20.\n\n**Reconciliation (2026-08-31):** `curator`, `mapping`, and `scheduling` existed\nin the tree without being added to this list when their epics landed. They are\nconfirmed as legitimate, narrowly-scoped domains under the same rules as the\noriginal six — each owns one capability family, declares typed contracts, and\nhas no cross-domain reach. This list was out of date, not the tree.\n\nConversely, `identity`, `agent`, `messaging`, and `plugin` are approved target\ndomains that do not yet exist as packages. Their behaviour currently lives in\npre-migration packages (see \"Complete package destination map\" below for the\nfull inventory). Do **not** create empty stub packages for them ahead of an\nactual migration PR — an unused stub is dead code and gives no test coverage.\nEach materialises when its corresponding migration decision below is resolved\nand implemented in the same change.\n\nMemory's final package placement remains undecided pending focused review of the\nexisting implementation.\n\nThe following are approved shared or composition packages, not domains:\n\n```text\ninternal/eventbus/\ninternal/policy/\ninternal/infrastructure/\ninternal/app/archied/\ninternal/app/agentworker/\n```\n\n## Remaining migration decisions\n\n### 1. Complete package destination map\n\nEvery current package MUST receive one explicit disposition:\n\n- move into a confirmed domain;\n- move under infrastructure as an implementation of a domain-owned contract;\n- move into application composition;\n- merge with an identified duplicate;\n- remain as an approved shared application contract;\n- move to a plugin, extra, deployment, example, or generated artifact;\n- be deleted after its replacement is proven.\n\nThe largest unreviewed areas include memory, tools and capabilities, workspace\nand worktree behaviour, gates, storage, events, web UI, containers, RPC\npackages, release management, and scheduling.\n\nThe destination matrix MUST name the behaviour owner, target path, dependencies\nto remove, state and contracts to preserve, migration prerequisites, and\ndeletion criteria.\n\n**`internal/webui` / UI Service — RATIFIED Phase 3 disposition (2026-09-07).**\nThe current owner is the in-process HTTP/SSE dashboard: SPA hosting, operator\nauthLine truncated
},
{
"url": "/dev/architecture/observability/",
@@ -130,7 +130,7 @@
"url": "/dev/development/frontend-ui/",
"title": "Frontend UI",
"section": "Development",
"body": "# Frontend UI\n\nAuthority: `.agents/skills/archie-web-ui` (stack facts, design-system rules,\ntone and accessibility). This page covers what a change must reach, not how it\nshould look.\n\n## Adding or changing a feature\n\n1. The feature lives in `ui/src/\u003cfeature\u003e/`. Move a helper to `ui/src/base/`\n only when a second feature needs it.\n2. Keep wire shapes, draft state and payload building in a plain `.ts` module\n beside the components (`ui/src/bindings/binding-draft.ts` is the pattern),\n and test it in `ui/test/*.test.ts`. Components stay thin.\n3. Every mutation goes through `api.*` in `ui/src/lib/api`, which sends the\n CSRF and Content-Type headers the server's mutation guard requires. A bare\n `fetch` is refused.\n4. A value the server also describes (task statuses, actions, workflow inputs)\n comes from the server. Keep a local default only so the first render works,\n and keep it in step (`ui/src/lib/task-meta.ts`).\n5. Run `task ui`, `task test:ui` and `cd ui \u0026\u0026 npm run typecheck`. Commit the\n regenerated `ui/dist/`; the Go binaries embed it. A fresh git worktree has\n no ignored `ui/node_modules`: run `cd ui \u0026\u0026 npm ci` before these commands.\n6. **One long-lived browser connection:** `ui/src/lib/api.ts` has the only\n `new EventSource`, owned by `stores/live-updates.ts`. Feature stores consume\n topics with `live.subscribe(topic, callback)`; do not open a page-owned SSE\n stream. Logs alone opt in while mounted, replacing that same EventSource\n with `?topics=logs\u0026since=\u003clast cursor\u003e`; a browser's `Last-Event-ID` only\n survives its own automatic reconnect, not a newly constructed EventSource.\n Task and log topics resume by cursor; control-plane, identity and apply-status\n topics replay the latest snapshot. Test a fetch and same-origin navigation\n after visiting Settings and Logs: the UI process serves HTTP/1.1 and six\n persistent streams can starve every later browser request.\n\n## Testing the dashboard\n\nAuthority: the runner itself is `ui/package.json`'s `test` script: `node --test\n--experimental-strip-types test/*.test.ts` over `ui/test/*.test.ts`. It runs\npure TypeScript modules with `node`'s type stripping — no SFC loader, no DOM,\nno second dependency beyond what the app already carries. That is the standard,\nand a decision record for when it is deliberately not enough:\n\n- **Test archie's logic in a module; leave SFC wiring alone.** Wire shapes,\n draft state, copy conditions, filters and payload building live in plain\n `.ts` files beside the components and are tested in `ui/test/`. Components\n stay thin, so there is normally nothing in an SFC that is archie logic.\n- **Installed primitives are upstream-tested, not re-tested here.**\n `reka-ui` (2.10.5 here) maintains colocated tests per component in its own\n repo (Vitest/jsdom + axe: Switch, TagsInput, NumberField, Select, the\n arrow-navigation composables). The npm package ships none of those files, so\n this coverage is inherited through the version bound, not the tarball —\n `grep -l test node_modules/reka-ui` finds nothing. Duplicating them as local\n mount tests would test a dependency.\n- **A `.vue` import cannot run under the module runner.** `node --test` fails\n such a module with `ERR_UNKNOWN_FILE_EXTENSION: .vue` and the runner has no\n built-in loader for SFCs. A file under `ui/test/` that imports a `.vue` is\n therefore a broken test, not a weak one.\n- **If a DOM mount harness is ever needed** (a page-level accessibility\n contract, cross-component interaction, focus behaviour): `happy-dom` +\n `@vue/test-utils` + `vitest` scoped to a separate DOM test glob, keeping the\n module runner for everything else. Measured on the ui tree (vitest 5.0.2,\n happy-dom 20.14.5, @vue/test-utils 2.5.1): they add 24 top-level\n `node_modules` entries and 27 MB on a 296 MB install, a working 2-test mount\n suite (role=radiogroup vs role=switch, aria-label, click activation against\n the real SFCs) runs in 1.2 s on top of the current 0.5 s module suite, and\n `vue-tsc` (7.4 s) dominates `task test:ui` either way. The single-runner\n alternative (absorbing the existing suite into vitest) was measured\n infeasible as-is: all 29 existing files import `node:test`, which vitest\n cannot collect (\"No test suite found in file\" for every one), so it would be\n a rewrite of the whole suite, and its runtime was not measured. Nothing in\n `ui/dist` or CI changes either way; happy-dom is pure JS, no browser binary.\n- **What the module standard does not verify:** whether a call site passes an\n `aria-label` or which `v-if` branch shows. Those live in SFC templates and\n are only visible with a mount. If a page's accessible-name wiring is the\n risk, say so in the page's task and reach for the harness decision above\n rather than quietly growing mount tests. Never suppress a lint rule or a\n test to keep Line truncated
"body": "# Frontend UI\n\nAuthority: `.agents/skills/archie-web-ui` (stack facts, design-system rules,\ntone and accessibility). This page covers what a change must reach, not how it\nshould look.\n\n## Adding or changing a feature\n\n1. The feature lives in `ui/src/\u003cfeature\u003e/`. Move a helper to `ui/src/base/`\n only when a second feature needs it.\n2. Keep wire shapes, draft state and payload building in a plain `.ts` module\n beside the components (`ui/src/bindings/binding-draft.ts` is the pattern),\n and test it in `ui/test/*.test.ts`. Components stay thin.\n3. Every mutation goes through `api.*` in `ui/src/lib/api`, which sends the\n CSRF and Content-Type headers the server's mutation guard requires. A bare\n `fetch` is refused.\n4. A value the server also describes (task statuses, actions, workflow inputs)\n comes from the server. Keep a local default only so the first render works,\n and keep it in step (`ui/src/lib/task-meta.ts`).\n5. Run `task ui`, `task test:ui` and `cd ui \u0026\u0026 npm run typecheck`. Commit the\n regenerated `ui/dist/`; the Go binaries embed it. A fresh git worktree has\n no ignored `ui/node_modules`: run `cd ui \u0026\u0026 npm ci` before these commands.\n6. **One long-lived browser connection:** `ui/src/lib/api.ts` has the only\n `new EventSource`, owned by `stores/live-updates.ts`. Feature stores consume\n topics with `live.subscribe(topic, callback)`; do not open a page-owned SSE\n stream. Logs alone opt in while mounted, replacing that same EventSource\n with `?topics=logs\u0026since=\u003clast cursor\u003e`; a browser's `Last-Event-ID` only\n survives its own automatic reconnect, not a newly constructed EventSource.\n Task and log topics resume by cursor; control-plane, identity and apply-status\n topics replay the latest snapshot. Test a fetch and same-origin navigation\n after visiting Settings and Logs: the UI process serves HTTP/1.1 and six\n persistent streams can starve every later browser request.\n\n## Testing the dashboard\n\nAuthority: the runner itself is `ui/package.json`'s `test` script: `node --test\n--experimental-strip-types test/*.test.ts` over `ui/test/*.test.ts`. It runs\npure TypeScript modules with `node`'s type stripping — no SFC loader, no DOM,\nno second dependency beyond what the app already carries. That is the standard,\nand a decision record for when it is deliberately not enough:\n\n- **Test archie's logic in a module; leave SFC wiring alone.** Wire shapes,\n draft state, copy conditions, filters and payload building live in plain\n `.ts` files beside the components and are tested in `ui/test/`. Components\n stay thin, so there is normally nothing in an SFC that is archie logic.\n- **A dashboard rule that mirrors a server rule is pinned by a shared\n fixture, not two independent tests.** `ui/src/settings/validation.ts`\n hand-mirrors three control-plane validators, and the model-role rule had\n already drifted because nothing tied the two sides together. Each case now\n lives once in `internal/app/controlplane/testdata/dashboard-validators.json`:\n a Go test in that package runs the real server validators over it, and\n `ui/test/control-plane-validator-parity.test.ts` runs the real dashboard\n validators over the same cases. Changing one rule alone fails the other\n side's test. Add a case to the fixture, never to one side only.\n- **Installed primitives are upstream-tested, not re-tested here.**\n `reka-ui` (2.10.5 here) maintains colocated tests per component in its own\n repo (Vitest/jsdom + axe: Switch, TagsInput, NumberField, Select, the\n arrow-navigation composables). The npm package ships none of those files, so\n this coverage is inherited through the version bound, not the tarball —\n `grep -l test node_modules/reka-ui` finds nothing. Duplicating them as local\n mount tests would test a dependency.\n- **A `.vue` import cannot run under the module runner.** `node --test` fails\n such a module with `ERR_UNKNOWN_FILE_EXTENSION: .vue` and the runner has no\n built-in loader for SFCs. A file under `ui/test/` that imports a `.vue` is\n therefore a broken test, not a weak one.\n- **If a DOM mount harness is ever needed** (a page-level accessibility\n contract, cross-component interaction, focus behaviour): `happy-dom` +\n `@vue/test-utils` + `vitest` scoped to a separate DOM test glob, keeping the\n module runner for everything else. Measured on the ui tree (vitest 5.0.2,\n happy-dom 20.14.5, @vue/test-utils 2.5.1): they add 24 top-level\n `node_modules` entries and 27 MB on a 296 MB install, a working 2-test mount\n suite (role=radiogroup vs role=switch, aria-label, click activation against\n the real SFCs) runs in 1.2 s on top of the current 0.5 s module suite, and\n `vue-tsc` (7.4 s) dominates `task test:ui` either way. The single-runner\n alternative (absorbing the existing suite into vitest) was measured\n infeasible as-is: all 29 existing files impLine truncated
},
{
"url": "/dev/development/merging-work/",
@@ -154,7 +154,7 @@
"url": "/dev/development/workflows/",
"title": "Workflows",
"section": "Development",
"body": "# Workflows\n\nAuthority: `docs/architecture/agent-system.md` for the lifecycle,\n`docs/prds/event-automation.md` for inputs, repository modes and profiles.\n\nWorkflow definitions are YAML stored as a control-plane resource. The State\nStore validates them and the agent compiles them, each against the step\nvocabulary it registers, so both must see the same step types.\n\n## Adding a step type\n\n1. Write the factory in `internal/domain/workflow` as a `StepType`. It decodes\n and validates its settings and returns an error for anything it cannot\n honour. The factory runs when a definition is saved and again when it is\n compiled, so a bad definition is refused on save.\n2. Register it through a provider in\n `internal/infrastructure/workflowsteps/workflowsteps.go`. That one list is\n what every process registers, which is how the validating and executing\n sides agree.\n3. An agent-driven step builds on `AgentStage`; do not start a second path to\n the agent runner.\n4. A step that needs no repository goes in `repoFreeStepTypes` in\n `definition.go`, or workflows declaring `repository: none` or `optional`\n cannot use it. A step that needs one may assume `tc.Dir` holds a worktree\n only when the workflow's repository is `required`.\n5. A step that ends the workflow sets `tc.Outcome`. A workflow that runs out of\n steps without an outcome is parked as a definition bug.\n6. Add a parse test (settings accepted and refused) and a run test with fake\n `Trees` and agent runner.\n\n## Changing the definition format\n\n- The fields a caller needs without the engine (inputs, repository, profile)\n live in `task.WorkflowInterface`, because the dashboard process must not\n import the workflow engine (`cmd/archie-ui/architecture_test.go`). Anything a\n binding or the dashboard checks goes there.\n- The YAML decoder rejects unknown keys. Existing stored definitions must keep\n parsing, so a new key is optional with a default that preserves today's\n behaviour.\n- Examples in `examples/workflows/` are parsed against the registered\n vocabulary by `TestExampleWorkflowsParse`. Add one for a new capability.\n\n## Enabling and disabling\n\nThe `workflow-enablement` control-plane resource lists, per org, the workflows\nthat org has disabled; anything unlisted is enabled. Its readers are the\ndaemon's binding dispatch (a disabled workflow's binding is skipped like an\nunarmed one), the dashboard's binding save (refused) and binding list\n(`workflow_disabled`), and work requests. A new reader of \"may this org run\nthis workflow\" goes through `task.WorkflowEnablement.Enabled`, never a second\nlist. The dashboard acts in the default org until access resolves a caller's\nown.\n\n## Starting a workflow from chat\n\n`task_spawn` accepts `workflow` and an `inputs` object. The gateway forwards\ninputs through the task creation contract; it does not interpret individual\nworkflows. The daemon checks inputs against the pinned `WorkflowInterface`\nbefore acquiring a container and parks invalid requests with the reason.\n"
"body": "# Workflows\n\nAuthority: `docs/architecture/agent-system.md` for the lifecycle,\n`docs/prds/event-automation.md` for inputs, repository modes and profiles.\n\nWorkflow definitions are YAML stored as a control-plane resource. The State\nStore validates them and the agent compiles them, each against the step\nvocabulary it registers, so both must see the same step types.\n\n## Adding a step type\n\n1. Write the factory in `internal/domain/workflow` as a `StepType`. It decodes\n and validates its settings and returns an error for anything it cannot\n honour. The factory runs when a definition is saved and again when it is\n compiled, so a bad definition is refused on save.\n2. Register it through a provider in\n `internal/infrastructure/workflowsteps/workflowsteps.go`. That one list is\n what every process registers, which is how the validating and executing\n sides agree.\n3. An agent-driven step builds on `AgentStage`; do not start a second path to\n the agent runner.\n4. A step that needs no repository goes in `repoFreeStepTypes` in\n `definition.go`, or workflows declaring `repository: none` or `optional`\n cannot use it. A step that needs one may assume `tc.Dir` holds a worktree\n only when the workflow's repository is `required`.\n5. A step that ends the workflow sets `tc.Outcome`. A workflow that runs out of\n steps without an outcome is parked as a definition bug.\n6. Add a parse test (settings accepted and refused) and a run test with fake\n `Trees` and agent runner.\n\n## Retiring a step type\n\nA step word cannot simply be deleted. A stored or pinned definition is\nvalidated as a whole collection, so one unknown word fails EVERY workflow's\ndefinition to decode and the workflow-definitions resource becomes unreadable.\nRetire a stage by removing it from its workflow's `Stages` and adding it to\n`retiredSteps` in `definition.go`, which keeps the word resolving as an inert\nstage. The retired stage must do nothing: the behaviour it used to perform now\nbelongs to whatever layer took it over (the remediate workflow's in-container\nresume moved to daemon worktree preparation, `archie-core-866m`; the workflow\nstill binds the daemon-prepared directory and the PR branch in a `prepare`\nstage before any agent work).\n\n## Changing the definition format\n\n- The fields a caller needs without the engine (inputs, repository, profile)\n live in `task.WorkflowInterface`, because the dashboard process must not\n import the workflow engine (`cmd/archie-ui/architecture_test.go`). Anything a\n binding or the dashboard checks goes there.\n- The YAML decoder rejects unknown keys. Existing stored definitions must keep\n parsing, so a new key is optional with a default that preserves today's\n behaviour.\n- A declared `outputs:` entry is written through a capture tool, so it forces\n the captures need: `task.WorkflowInterface.Needs` resolves that from the\n declaration rather than the author also writing `needs.captures`. A Kit\n profile whose harness serves no capture tools is refused before dispatch.\n- Examples in `examples/workflows/` are parsed against the registered\n vocabulary by `TestExampleWorkflowsParse`. Add one for a new capability.\n\n## Enabling and disabling\n\nThe `workflow-enablement` control-plane resource lists, per org, the workflows\nthat org has disabled; anything unlisted is enabled. Its readers are the\ndaemon's binding dispatch (a disabled workflow's binding is skipped like an\nunarmed one), the dashboard's binding save (refused) and binding list\n(`workflow_disabled`), and work requests. A new reader of \"may this org run\nthis workflow\" goes through `task.WorkflowEnablement.Enabled`, never a second\nlist. The dashboard acts in the default org until access resolves a caller's\nown.\n\n## Starting a workflow from chat\n\n`task_spawn` accepts `workflow` and an `inputs` object. The gateway forwards\ninputs through the task creation contract; it does not interpret individual\nworkflows. The daemon checks inputs against the pinned `WorkflowInterface`\nbefore acquiring a container and parks invalid requests with the reason.\n"
},
{
"url": "/dev/prds/01-project-management/",
@@ -352,7 +352,7 @@
"url": "/dev/prds/mcp-tool-completion/",
"title": "MCP client and tool-guardrail completion",
"section": "PRDs",
"body": "# MCP client and tool-guardrail completion\n\n**Status:** Approved\n\nEpic: `archie-core-1786637493439-140-2cee2e9b` / GitHub `#161` (\"[EPIC] Tools:\ncentral registry, MCP client, guardrails\").\n\nThis document is scoped to five sub-features of that epic: MCP\nserver-initiated sampling, the per-server parallel-tool-calls flag, sandboxed\nMCP tool servers, a Firecracker sandbox backend, and sequential dispatch with\nper-result previews.\n\nOut of scope, and not to be touched: the central registry\n(`internal/tools/registry.go`), all three MCP transports (stdio/HTTP/SSE,\n`internal/tools/mcp/transport*.go`), reconnect-with-backoff\n(`transport.go`), guardrail loop detection\n(`internal/tools/guardrail.go`), idempotent/mutating classification\n(`internal/tools/classification.go`), concurrent tool dispatch (`#183`), and\nprogressive disclosure wiring (`internal/app/archied`). These are settled and\nproduction-wired.\n\n## Problem\n\n`bd search \"MCP client\"` shows five closed transport beads and one open:\nserver-initiated sampling (`#151`). `internal/tools/mcp/client.go`\nholds a single `callMu *sync.Mutex` that serialises every `tools/call`\nregardless of server — `ParallelToolCalls` does not exist on the `MCPServer`\nconfig struct (`internal/config/config.go`), confirming `#177` is a\nreal gap, not a stale bead. `#178` (wire MCP tool servers for the chat agent,\nsandboxed) and `#179` (Firecracker substrate) have no corresponding code —\nMCP servers run as whatever the config's `Command`/`Args` launches,\nunsandboxed. `#182` (sequential tool dispatch with per-result previews) is\nopen; concurrent dispatch (`#183`) is closed, so there are two dispatch paths\nand this PRD decides how they coexist.\n\n## Design\n\nSub-features are independent; there is no new shared type this document\nintroduces. Each section below stands alone.\n\n## MCP server-initiated sampling (`#151`)\n\n**Decision.** Implement as a callback on `mcp.Client`, not a second\ntransport. The MCP spec's `sampling/createMessage` request flows server→client\nover the same connection each transport already maintains\n(`transport.go`'s stdio/HTTP/SSE all multiplex request and response frames).\nAdd a `SamplingHandler func(context.Context, SamplingRequest) (SamplingResult, error)`\nfield to the client's config struct next to the existing transport fields,\ndefaulting to nil (server sampling requests get a JSON-RPC \"not supported\"\nerror when unset — never a silent drop).\n\n**Call site.** `internal/tools/mcp/client.go` — add the handler field and a\ndispatch branch in whatever function demuxes incoming server\nmessages by method name (the same switch that already routes\n`notifications/*`). Wire a concrete handler in\n`internal/app/archied/main.go` that routes the sampling request through the\nsame chat-model call path `chatGenerateOptions` already uses, so a\nserver-initiated sample gets the daemon's configured model, not a second\nprovider client.\n\n**Reuse check.** The chat model invocation path already exists\n(`chatGenerateOptions`, `main.go`); this reuses it rather than adding a\nparallel model-calling path for MCP sampling.\n\n**Failure mapping.**\n\n| condition | contract error |\n| ------------------------------- | ------------------------------------------------------------------------------------------ |\n| no `SamplingHandler` configured | JSON-RPC method-not-found, per spec |\n| handler returns error | JSON-RPC internal error, message from handler |\n| handler timeout | JSON-RPC timeout error; use the transport's existing request-timeout config, not a new one |\n\n**Testing.** Fakeable entirely — a stub `SamplingHandler` and a fake\ntransport that emits a `sampling/createMessage` frame. No real backend\nneeded; belongs in `task check`.\n\n## Per-server parallel-tool-calls flag (`#177`)\n\n**Decision.** `ParallelToolCalls bool` (default `false`) goes on `MCPServer`\nin `internal/config/config.go`, next to `Transport`/`Command`. The\nmutex in `client.go` becomes per-server: when `ParallelToolCalls` is\nfalse (default, and the only behaviour), keep the single `callMu`\nserialising that server's calls exactly as now; when true, drop the mutex for\nthat server's `Client` instance entirely and let the caller's own\nconcurrency (already proven by the closed concurrent-dispatch bead, `#183`)\nhandle it. This is additive — no server's behaviour changes unless an operator\nopts in.\n\n**Call site.** `internal/config/config.go` (struct field),\n`internal/tools/mcp/client.go` (make the mutex conditional on the\nfield), and wherever `MCPServer` configs are turned into `Client` instances\nat daemon startup (the constructor that reads `internal/config`'s MCP block).\n`config.example.toml`'s commenLine truncated
"body": "# MCP client and tool-guardrail completion\n\n**Status:** Approved\n\nEpic: `archie-core-1786637493439-140-2cee2e9b` / GitHub `#161` (\"[EPIC] Tools:\ncentral registry, MCP client, guardrails\").\n\nThis document is scoped to five sub-features of that epic: MCP\nserver-initiated sampling, the per-server parallel-tool-calls flag, sandboxed\nMCP tool servers, a Firecracker sandbox backend, and sequential dispatch with\nper-result previews.\n\nOut of scope, and not to be touched: the central registry\n(`internal/tools/registry.go`), all three MCP transports (stdio/HTTP/SSE,\n`internal/tools/mcp/transport*.go`), reconnect-with-backoff\n(`transport.go`), guardrail loop detection\n(`internal/tools/guardrail.go`), idempotent/mutating classification\n(`internal/tools/classification.go`), concurrent tool dispatch (`#183`), and\nprogressive disclosure wiring (`internal/app/archied`). These are settled and\nproduction-wired.\n\n## Problem\n\n`bd search \"MCP client\"` shows five closed transport beads and one open:\nserver-initiated sampling (`#151`). `internal/tools/mcp/client.go`\nholds a single `callMu *sync.Mutex` that serialises every `tools/call`\nregardless of server — `ParallelToolCalls` does not exist on the `MCPServer`\nconfig struct (`internal/config/config.go`), confirming `#177` is a\nreal gap, not a stale bead. `#178` (wire MCP tool servers for the chat agent,\nsandboxed) and `#179` (Firecracker substrate) have no corresponding code —\nMCP servers run as whatever the config's `Command`/`Args` launches,\nunsandboxed. `#182` (sequential tool dispatch with per-result previews) is\nopen; concurrent dispatch (`#183`) is closed, so there are two dispatch paths\nand this PRD decides how they coexist.\n\n## Design\n\nSub-features are independent; there is no new shared type this document\nintroduces. Each section below stands alone.\n\n## MCP server-initiated sampling (`#151`)\n\n**Decision.** Implement as a callback on `mcp.Client`, not a second\ntransport. The MCP spec's `sampling/createMessage` request flows server→client\nover the same connection each transport already maintains\n(`transport.go`'s stdio/HTTP/SSE all multiplex request and response frames).\nAdd a `SamplingHandler func(context.Context, SamplingRequest) (SamplingResult, error)`\nfield to the client's config struct next to the existing transport fields,\ndefaulting to nil (server sampling requests get a JSON-RPC \"not supported\"\nerror when unset — never a silent drop).\n\n**Call site.** `internal/tools/mcp/client.go` — add the handler field and a\ndispatch branch in whatever function demuxes incoming server\nmessages by method name (the same switch that already routes\n`notifications/*`). Wire a concrete handler in\n`internal/app/archied/main.go` that routes the sampling request through the\nsame chat-model call path `chatGenerateOptions` already uses, so a\nserver-initiated sample gets the daemon's configured model, not a second\nprovider client. The agent worker hosts the same configured MCP servers on the\ntask path and runs its own model runtime, so `internal/app/agentworker` wires\nthe same seam from the task's builder model: a server that samples succeeds\ninside a chat turn and inside a task, never one and not the other. The message\nmapping and the token bound are shared (`internal/agentexec`, next to the\nruntime both paths call models through) so both paths bound a sample\nidentically.\n\n**Reuse check.** The chat model invocation path already exists\n(`chatGenerateOptions`, `main.go`) and the task model path already exists\n(`agentexec.NewRuntime`, the runtime an agent stage runs on); this reuses both\nrather than adding a parallel model-calling path for MCP sampling.\n\n**Failure mapping.**\n\n| condition | contract error |\n| ------------------------------- | ------------------------------------------------------------------------------------------ |\n| no `SamplingHandler` configured | JSON-RPC method-not-found, per spec |\n| handler returns error | JSON-RPC internal error, message from handler |\n| handler timeout | JSON-RPC timeout error; use the transport's existing request-timeout config, not a new one |\n\n**Testing.** Fakeable entirely — a stub `SamplingHandler` and a fake\ntransport that emits a `sampling/createMessage` frame. No real backend\nneeded; belongs in `task check`.\n\n## Per-server parallel-tool-calls flag (`#177`)\n\n**Decision.** `ParallelToolCalls bool` (default `false`) goes on `MCPServer`\nin `internal/config/config.go`, next to `Transport`/`Command`. The\nmutex in `client.go` becomes per-server: when `ParallelToolCalls` is\nfalse (default, and the only behaviour), keep the single `callMu`\nserialising that server's calls exactly as now; when true, drop the muLine truncated
},
{
"url": "/dev/prds/memory-engine-unification/",
@@ -412,7 +412,7 @@
"url": "/dev/prds/plugin-settings-live/",
"title": "Plugin settings reconcile live",
"section": "PRDs",
"body": "# Plugin settings reconcile live\n\n**Status:** Draft\n**Tracking:** GitHub [#1154](https://github.com/samcharles93/archie-core/issues/1154), epic [#998](https://github.com/samcharles93/archie-core/issues/998)\n**Authority:** `docs/prds/runtime-control-plane.md` decides each feature owns how\nits changes take effect; `docs/architecture/plugins-and-extensions.md` owns the\nplugin engine rule; `docs/prds/control-plane-apply-status.md` owns where apply\nreports go.\n\n## Outcome\n\nAn operator drops a plugin, module or secret-engine file into the configured\ndirectory while archied runs, and the daemon loads it without a restart. An\nedit to an installed file replaces what is running. Deleting an installed file\nnever stops running code; apply status says so. A stored change to\n`plugin-settings` applies live for additions instead of requiring a restart.\n\n## Triggers\n\n`plugin-settings` joins the kinds archied watches (`PluginSettingsKind` moves\nto the live list in `internal/app/archied/control_plane.go`). After the daemon\nlayers a new version it reconciles the three directories the document names.\nA second trigger needs no document change: a poller reconciles on an interval,\nso a file dropped into a directory loads on its own. The poll interval is the\napply-status restamp constant; no filesystem-watch dependency is added. The\npoller reads the current document's directories from the running config each\ntick, so a changed `plugin_dir`, `module_dir` or `secret_engine_dir` retargets\nreconciliation without a restart.\n\n## Rules\n\n- Adding and changing load; removing never unloads. Yaegi interpreters are not\n reclaimed while a caller still exists and there is no unload path; the\n running set can only grow.\n- Each file is loaded through a fresh interpreter (the boot pattern). A changed\n file therefore loads cleanly, and replacement switches consumers to the new\n interpretation; files are content-hashed so only new and changed files load.\n- A file that fails to load is skipped and reported; already-running code keeps\n running. This is the degrade-and-skip convention `plugin.LoadDir` and\n `Registry.LoadDir` already apply at boot. Boot keeps its reject-at-load rule\n for the module directory; a broken file at runtime skips without aborting.\n- A directory that does not exist reconciles to an empty set.\n\n## The three registries\n\n- **Secret engines** (`internal/secret`). Reconciliation re-runs\n `Registry.LoadDir` against the live directory. `Register` replaces by name,\n so an edited engine takes effect for the next resolution. A deleted engine\n stays registered: existing references keep resolving through it until\n restart, and the record says so. Engines are stateless resolvers, so a swap\n needs no lifecycle step.\n- **Modules** (`internal/domain/eda/module`): reconciliation re-runs\n `ModuleRegistry.Register` for the kind's file. The invoker maps\n last-write-wins, and playbook dispatch resolves through the registry at\n invoke time, so the swap reaches a running playbook store unchanged. A file\n deleted for a kind leaves the old invoker in place, running; reported.\n- **Plugins** (`internal/plugin`): `Host` grows running-state operations beside\n `Register`, `Start` and `Stop`. One adds a single module while the host is\n running: it is inserted in dependency order, started alone, and on failure\n rolled back alone, leaving every previously started module running. The\n other replaces a module whose manifest id matches: stop it, start the new\n one, and restart the old one when the new start fails. Replacing a module\n that another depends on is refused, not swapped. These operations obey the\n plugin engine rule: the host still invokes start and stop itself, isolates\n the failure to the one module, and keeps a rollback. A deleted file makes no\n host call: the module runs on and is reported.\n\n## `skills_dir`\n\n`skills_dir` is not part of this change. The daemon refuses a candidate\n`plugin-settings` document whose `skills_dir` differs from the value it is\nrunning, reporting that the field applies on restart, until a skill\nreconciliation exists to carry it. The refusal is the existing rejected-edit\napply-status shape: the store keeps the version, the document is not applied,\nlast-known-good keeps running.\n\n## Apply status\n\nOne record: `archied` on `plugin-settings`. The reconciliation outcome is\nreported against the version the process is running. The record's error field\ncarries the outstanding reconciliation problem: a deleted file still running, a\nfile that failed to load, or the directory failing to read. A reconciliation\nwith nothing outstanding clears the error. The version stays the version the\nprocess is running; an add that succeeded is live even while a removal is\noutstanding, and the error says so rather than showing the kind as\nrestart-required. archie-gateway keeps boot-time loading; its record simply\nstays on the vLine truncated
"body": "# Plugin settings reconcile live\n\n**Status:** Draft\n**Revision:** 2\n**Tracking:** GitHub [#1154](https://github.com/samcharles93/archie-core/issues/1154), epic [#998](https://github.com/samcharles93/archie-core/issues/998)\n**Authority:** `docs/prds/runtime-control-plane.md` decides each feature owns how\nits changes take effect; `docs/architecture/plugins-and-extensions.md` owns the\nplugin engine rule; `docs/prds/control-plane-apply-status.md` owns where apply\nreports go.\n\n## Outcome\n\nAn operator drops a plugin, module or secret-engine file into a configured\ndirectory while archied runs, and the daemon loads it without a restart. An\nedit to an installed file replaces what is running. A delete never stops running\ncode; apply status says so. A stored `plugin-settings` change applies live for\n`plugin_dir`, `module_dir` and `secret_engine_dir`; a change to `skills_dir` is\nrefused.\n\n## Triggers\n\n`PluginSettingsKind` joins `runtimeResourceKinds` in\n`internal/app/archied/control_plane.go`, and its descriptor `ApplyMode` becomes\n`live` in `internal/app/controlplane/operational_settings.go`.\n\nA poller reconciles on the apply-status restamp interval, so a dropped file\nloads without a filesystem watch. Each tick re-runs both halves the boot path\nruns, against the directories in the running config: the directory loads\n(`secret.Registry.LoadDir`, `ModuleRegistry.Register`, `plugin.LoadDir`) and the\nlayering (`boot.runtimeConfig`), which re-resolves provider credentials. A\nprovider disabled at boot for an unresolvable engine is re-resolved on the first\ntick that finds it; a changed directory retargets the next tick without a\nrestart.\n\n## Rules\n\n- Adding and changing load; removing never unloads. Yaegi interpreters are not\n reclaimed while a caller still exists and there is no unload path; the running\n set can only grow. A removal stays outstanding in the apply-status error until\n the process restarts.\n- Each file is loaded through a fresh interpreter (the boot pattern) and\n content-hashed, so only new and changed files load.\n- A file that fails to load is skipped and reported; already-running code keeps\n running, the degrade-and-skip convention `plugin.LoadDir` and\n `Registry.LoadDir` already apply at boot. Boot keeps its reject-at-load rule\n for the module directory; a broken file at runtime skips without aborting.\n- A directory that does not exist reconciles to an empty set.\n\n## The three registries\n\n- **Secret engines** (`internal/secret`). Reconciliation re-runs\n `Registry.LoadDir` against the live directory. `Register` replaces by name, so\n an edited engine takes effect for the next resolution and a dropped engine\n resolves on the next tick. A deleted engine stays registered: existing\n references keep resolving through it until restart. Engines are stateless\n resolvers, so a swap needs no lifecycle step.\n- **Modules** (`internal/domain/eda/module`). Reconciliation re-runs\n `ModuleRegistry.Register` for the kind's file. Playbook dispatch resolves\n through the registry at invoke time, so the swap reaches a running playbook\n store unchanged; a deleted file leaves the old invoker in place, running.\n- **Plugins** (`internal/plugin`). `Host` grows running-state operations beside\n `Register`, `Start` and `Stop`, and `Register` keeps its rule that\n registration is allowed only before the first `Start`, so these are new host\n methods rather than a relaxed `Register`. One adds a single module while the\n host is running: inserted in dependency order, started alone, and rolled back\n alone on failure, leaving every previously started module running. The other\n replaces a module whose manifest id matches: stop it, start the new one, and\n restart the old one when the new start fails; replacing a module another\n running module depends on is refused. A module's `Start` is its only check\n (`plugin.Module` has no probe, and one manifest id cannot be registered\n twice), so a replacement checks by starting and keeps a rollback rather than\n switching first. These operations obey the plugin engine rule: the host\n invokes start and stop itself and isolates the failure to one module. A\n deleted file makes no host call: the module runs on.\n\n## `skills_dir`\n\n`skills_dir` is not part of this change. The daemon refuses a `plugin-settings`\ndocument whose `skills_dir` differs from the value in force. The check is a\npre-layer comparison on the live path (`boot.applyRuntimeResourceUpdate` in\n`internal/app/archied/control_plane.go`), not an addition to\n`validatePluginSettings` in `internal/app/controlplane/operational_settings.go`,\nwhich stays value-agnostic per the #1143 decision recorded there: a\nchange-immutability rule needs the running config, and #1143's validates-nothing\nstill holds for values. The write lands, the store keeps the version, the daemon\npublishes nothing, the refusal is reported against that stored versLine truncated
},
{
"url": "/dev/prds/pr-review-agent/",
@@ -424,7 +424,7 @@
"url": "/dev/prds/pr-review-operator-response/",
"title": "PR review: the operator's answer to the approval gate",
"section": "PRDs",
"body": "# PR review: the operator's answer to the approval gate\n\n**Status:** Draft\n**Authority:** `docs/prds/pr-review-agent.md` (\"Operator approval\") for the gate, and `docs/prds/state-store-contract.md` (rev. 2e) for every task field that crosses the State Store.\n**Compounds with:** `docs/prds/execution-tree-state-machine.md` (a run cannot resume partway through its stage list).\n\n## Decision 1 — where the operator answers\n\nThe dashboard's `POST /api/tasks/{id}/action` (`internal/webui/api_tasks.go`, `handleTaskAction`) is the authoritative surface: the only one that resolves the actor from a credential and can carry a payload. Chat is a shortcut, not a second authority — `/approve` (`internal/gateway/gateway.go`, `Router.handleApprove`) answers the post decision with every offered finding selected, and gains no instruction or selection syntax.\n\n`taskstate.Action` gains `rereview`, and `taskstate.Actions` returns `approve`, `reject` and `rereview` for `WaitingHuman`. `approve` posts the review, `reject` discards it through the existing terminal decline, `rereview` reruns the review phases with instructions. The action body carries `instructions`, required by `rereview`, and `findings`, the finding keys `approve` posts; an absent selection means all of them.\n\nThe path is `handleTaskAction` → `applyOperatorTaskAction` → `Chat.Contract.ApplyOperatorTaskAction` → the daemon's `LocalChatAdapter` → `taskactions.Client` (NATS `archie.gateway.task-action`) → `taskactions.Service.Apply` → one guarded store write that records the decision and requeues. Chat's `/approve` path (`StoreTaskController.Approve` → `chatTaskControllerAdapter.ApproveChatTask`) must reach that same write through the same function, so the two surfaces cannot record different decisions for one operator intent.\n\nA requeue resumes the workflow the waiting state names, never a name the handler hardcodes. `taskactions.Service.apply` and `chatTaskControllerAdapter.ApproveChatTask` requeue every approval under `implement`, which is the feasibility handoff: an approval whose wait recorded `pr-review` would run the implement workflow. Feasibility must keep that handoff by naming `implement` on the task before it enters the wait, which is the invariant the handler needs — the workflow a task carries is the workflow it runs next. `Requeue` treats an empty workflow as \"keep the task's\".\n\n### The review the operator answers\n\nThe gate writes the review it is holding onto the task before it ends the run: one JSON column, `review_gate`, a sibling of `review_payload`, holding the scored findings, the head SHA, the pull request's identity and the workflow the wait resumes. The run's existing `Store.Update` persists it. The same column carries the answer back: the response path's guarded write fills in the outcome, the selection and the instructions, and clears the offer for a re-review.\n\nThe operator reads the offer through the existing task read — `taskView.Task` on `GET /api/tasks` and `GET /api/tasks/{id}/debug` — so no new read endpoint is needed.\n\nThis record is load-bearing because the run that computed the findings ends at the gate and the engine cannot resume it partway. Without it, \"post the selected findings\" cannot mean the findings the operator saw: the resumed run recomputes them with agent calls. An `approve` resume therefore posts the recorded review filtered by the selection, and the review the resumed phases recompute never reaches the output.\n\n## Decision 2 — the cap\n\nThe cap is a column: `rereview_rounds` on `tasks`, `bigint NOT NULL DEFAULT 0`, read by the response path and incremented in the same guarded write that requeues a re-review. A re-review at the cap is refused with the conflict sentinel the dashboard answers 409 with, and nothing is requeued. Two is a constant in the domain, not a config knob: the PRD fixes it.\n\n`Task.Attempt` cannot carry the count, because it counts every requeue. The two alternatives are worse.\n\n- The execution tree measures passes through the gate, not operator requests. A crash recovery (`RecoverStale`) or an operator retry re-runs the pipeline to the gate and would consume the cap; and a stage has no step read (`task.Store` carries none), so this route needs a new interface method anyway.\n- `Task.Notes` is the agent memory ledger, not scratch. `AgentStage.buildRequest` passes it into every generic agent request, `agentexec` renders it as a prompt section, and `persistAppendedNotes` appends to it. A counter parsed out of a field agents write to is the encoded-state defect this repository rejects, and instructions parked there would reach every unrelated workflow's agents.\n\n`RemediationRounds` is the precedent: one writer, one reader, one meaning, kept separate from `RetryCount` so unrelated retries cannot draw a budget down.\n\nThe write crosses the State Store. A goose migration adds the column; `internal/infrastructure/postgres/queriesLine truncated
"body": "# PR review: the operator's answer to the approval gate\n\n**Status:** Draft\n**Authority:** `docs/prds/pr-review-agent.md` (\"Operator approval\") for the gate, and `docs/prds/state-store-contract.md` (rev. 2e) for every task field that crosses the State Store.\n**Compounds with:** `docs/prds/execution-tree-state-machine.md` (a run cannot resume partway through its stage list).\n\n## Decision 1 — where the operator answers\n\nThe dashboard's `POST /api/tasks/{id}/action` (`internal/webui/api_tasks.go`, `handleTaskAction`) is the authoritative surface: the only one that resolves the actor from a credential and can carry a payload. Chat is a shortcut, not a second authority — `/approve` (`internal/gateway/gateway.go`, `Router.handleApprove`) answers the post decision with every offered finding selected, and gains no instruction or selection syntax.\n\n`taskstate.Action` gains `rereview`, and `taskstate.Actions` returns `approve`, `reject` and `rereview` for `WaitingHuman`. `approve` posts the review, `reject` discards it through the existing terminal decline, `rereview` reruns the review phases with instructions. The action body carries `instructions`, required by `rereview`, and `findings`, the finding keys `approve` posts; an absent selection means all of them.\n\nThe path is `handleTaskAction` → `applyOperatorTaskAction` → `Chat.Contract.ApplyOperatorTaskAction` → the daemon's `LocalChatAdapter` → `taskactions.Client` (NATS `archie.gateway.task-action`) → `taskactions.Service.Apply` → one guarded store write that records the decision and requeues. Chat's `/approve` path (`StoreTaskController.Approve` → `chatTaskControllerAdapter.ApproveChatTask`) must reach that same write through the same function, so the two surfaces cannot record different decisions for one operator intent.\n\nA requeue resumes the workflow the waiting state names, never a name the handler hardcodes. `taskactions.Service.apply` and `chatTaskControllerAdapter.ApproveChatTask` requeue with an empty workflow, which `Requeue` reads as \"keep the task's\", so neither surface hardcodes a workflow name: an approval whose wait recorded `pr-review` resumes `pr-review`. The feasibility handoff therefore has to name `implement` on the task before it enters the wait, which is the invariant the handler relies on — the workflow a task carries is the workflow it runs next. `Requeue` treats an empty workflow as \"keep the task's\".\n\n### The review the operator answers\n\nThe gate writes the review it is holding onto the task before it ends the run: one JSON column, `review_gate`, a sibling of `review_payload`, holding the scored findings, the head SHA, the pull request's identity and the workflow the wait resumes. The run's existing `Store.Update` persists it. The same column carries the answer back: the response path's guarded write fills in the outcome, the selection and the instructions, and clears the offer for a re-review.\n\nThe operator reads the offer through the existing task read — `taskView.Task` on `GET /api/tasks` and `GET /api/tasks/{id}/debug` — so no new read endpoint is needed.\n\nThis record is load-bearing because the run that computed the findings ends at the gate and the engine cannot resume it partway. Without it, \"post the selected findings\" cannot mean the findings the operator saw: the resumed run recomputes them with agent calls. An `approve` resume therefore posts the recorded review filtered by the selection, and the review the resumed phases recompute never reaches the output.\n\n## Decision 2 — the cap\n\nThe cap is a column: `rereview_rounds` on `tasks`, `bigint NOT NULL DEFAULT 0`, read by the response path and incremented in the same guarded write that requeues a re-review. A re-review at the cap is refused with the conflict sentinel the dashboard answers 409 with, and nothing is requeued. Two is a constant in the domain, not a config knob: the PRD fixes it.\n\n`Task.Attempt` cannot carry the count, because it counts every requeue. The two alternatives are worse.\n\n- The execution tree measures passes through the gate, not operator requests. A crash recovery (`RecoverStale`) or an operator retry re-runs the pipeline to the gate and would consume the cap; and a stage has no step read (`task.Store` carries none), so this route needs a new interface method anyway.\n- `Task.Notes` is the agent memory ledger, not scratch. `AgentStage.buildRequest` passes it into every generic agent request, `agentexec` renders it as a prompt section, and `persistAppendedNotes` appends to it. A counter parsed out of a field agents write to is the encoded-state defect this repository rejects, and instructions parked there would reach every unrelated workflow's agents.\n\n`RemediationRounds` is the precedent: one writer, one reader, one meaning, kept separate from `RetryCount` so unrelated retries cannot draw a budget down.\n\nThe write crosses the State Store. A goose migration adds the column; `Line truncated
},
{
"url": "/dev/prds/pr-review-remediation/",
@@ -466,7 +466,7 @@
"url": "/dev/prds/runtime-control-plane/",
"title": "Runtime control plane",
"section": "PRDs",
"body": "# Runtime control plane\n\n**Status:** Approved \n**Date:** 2026-09-20 \n**Tracking:** `archie-core-j28m`, `archie-core-1786637498420-327-bacadaae`\n\n## Outcome\n\nArchie can be managed from the Web UI or a messaging channel. Both use the same\nAPI.\n\n## Model\n\nEach feature owns its settings and commands. The database stores them. There is\nno global configuration document.\n\nThis includes:\n\n- workflows;\n- identities and bindings;\n- providers and models;\n- repository policies;\n- channels and schedules;\n- capture mappings and bindings;\n- tool settings;\n- prompts and personas;\n- plugin settings.\n\nPlugin code remains installed code. Only its settings are stored in the\ndatabase.\n\nEach feature defines its data, validation, permissions, editor fields, and how\nchanges take effect. Each feature also defines its allowed actions. For example,\nidentities can be created, suspended, and retired.\n\nRepository and identity settings remain separate. They are not stored as large\nrecords that unrelated features must share.\n\nThe State Store process hosts the control-plane API on its existing gRPC server.\nNo new process is added. Feature code validates changes. Storage code only\nwrites versions and audit records.\n\nEvery write records who made it, where it came from, a request ID, and the\nversion being edited. The API layer sets the actor; the request body cannot.\nThe server validates the change, then writes it with its audit record. It rejects\nedits to an old version.\n\nThe request ID is the write's idempotency key, and it is scoped per org and\nper kind, never global. The key is `(org_id, kind, request_id)` — the unique\nindex `idx_resource_history_org_kind_request` (migration `0021_resources_per_org.sql`)\n— so replaying a request ID is the same org's prior write of the same kind, and\na request ID reused across orgs or across kinds is a separate write. This is the\nsame scoping as the task idempotency key `org/identity/owner/repo/number`\n(`docs/prds/orgs-and-access.md`).\n\n## API\n\nThe API has four operations:\n\n1. **Catalogue** tells clients what can be viewed or changed and how to edit it.\n2. **Query** reads data and versions without exposing secrets.\n3. **Command** asks the owning feature to make a change.\n4. **Watch** streams changes and errors to clients.\n\nComplex features can use custom screens. They still use the same commands.\n\nA dashboard session or channel sender must map to an Archie identity before it\ncan run a command. Messaging syntax maps to the same commands as the Web UI.\n\n`archie-ui` receives only an API client. It does not receive a database handle,\ndaemon pointer, global configuration object, or secret value.\n\nThe database stores the requested settings. Each process reports the version it\nis using or why it failed. The UI shows both.\n\nSome changes require a restart. The UI shows them as pending. When a live change\nreplaces a running component, Archie starts and checks the new one before\nswitching. If the check fails, the old one keeps running.\n\n## Bootstrap, migration, and recovery\n\nBootstrap contains only the database path, listen addresses, credentials, TLS,\nand encryption-key references. State Store and control-plane RPCs use the same\nendpoint.\n\nWhen upgraded, existing instances are migrated to the database. Existing\ndatabase values are not overwritten, with one exception: a resource whose\ndocument the build ships is refreshed when the seed itself wrote the stored\ncopy, because that copy is a previous release's document rather than an\noperator's. After migration, settings in TOML are ignored and cannot block\nState Store startup.\n\nThe Web UI remains available when `archied` or Gateway is down. It shows failed\nchanges and can correct or roll them back.\n\nOffline commands can back up, restore, validate, and roll back the database.\nRecovery does not require `archied` or the Web UI. The database stores secret\nreferences, not secret values.\n\n## Workflows\n\nUsers create and edit workflows in the Web UI. Workflows are stored as versioned\nYAML. Built-in workflows can be overridden and restored.\n\nA workflow is a list of steps. Each step has a known type and its own settings.\nPlugins can add step types. Each run keeps the workflow version it started with.\n\nYaegi, `.archie/stages/*.go`, and `.archie/gate.go` will be removed. Their useful\nbehaviour must first be available as normal workflow steps.\n\n## First implementation\n\nThe first implementation makes task limits editable: maximum model steps,\nmaximum runtime, and maximum consecutive gate failures.\n\nIt includes:\n\n- validated task-limit fields;\n- version history and audit records;\n- the four API operations;\n- authenticated Web UI and messaging access;\n- live updates and apply status;\n- a Pinia-backed editor in the Web UI;\n- a messaging command using the same API;\n- migration of existing `[budgets]` values;\n- an end-to-end restart test.\n\nRuLine truncated
"body": "# Runtime control plane\n\n**Status:** Approved \n**Date:** 2026-09-20 \n**Tracking:** `archie-core-j28m`, `archie-core-1786637498420-327-bacadaae`\n\n## Outcome\n\nArchie can be managed from the Web UI or a messaging channel. Both use the same\nAPI.\n\n## Model\n\nEach feature owns its settings and commands. The database stores them. There is\nno global configuration document.\n\nThis includes:\n\n- workflows;\n- identities and bindings;\n- providers and models;\n- repository policies;\n- channels and schedules;\n- capture mappings and bindings;\n- tool settings;\n- prompts and personas;\n- plugin settings.\n\nPlugin code remains installed code. Only its settings are stored in the\ndatabase.\n\nEach feature defines its data, validation, permissions, editor fields, and how\nchanges take effect. Each feature also defines its allowed actions. For example,\nidentities can be created, suspended, and retired.\n\nRepository and identity settings remain separate. They are not stored as large\nrecords that unrelated features must share.\n\nThe State Store process hosts the control-plane API on its existing gRPC server.\nNo new process is added. Feature code validates changes. Storage code only\nwrites versions and audit records.\n\nEvery write records who made it, where it came from, a request ID, and the\nversion being edited. The API layer sets the actor; the request body cannot.\nThe server validates the change, then writes it with its audit record. It rejects\nedits to an old version.\n\nThe request ID is the write's idempotency key, and it is scoped per org and\nper kind, never global. The key is `(org_id, kind, request_id)` — the unique\nindex `idx_resource_history_org_kind_request` (migration `0021_resources_per_org.sql`)\n— so replaying a request ID is the same org's prior write of the same kind, and\na request ID reused across orgs or across kinds is a separate write. This is the\nsame scoping as the task idempotency key `org/identity/owner/repo/number`\n(`docs/prds/orgs-and-access.md`).\n\n## API\n\nThe API has four operations:\n\n1. **Catalogue** tells clients what can be viewed or changed and how to edit it.\n2. **Query** reads data and versions without exposing secrets.\n3. **Command** asks the owning feature to make a change.\n4. **Watch** streams changes and errors to clients.\n\nComplex features can use custom screens. They still use the same commands.\n\nA dashboard session or channel sender must map to an Archie identity before it\ncan run a command. Messaging syntax maps to the same commands as the Web UI.\n\n`archie-ui` receives only an API client. It does not receive a database handle,\ndaemon pointer, global configuration object, or secret value.\n\nThe database stores the requested settings. Each process reports the version it\nis using or why it failed. The UI shows both.\n\nSome changes require a restart. The UI shows them as pending. When a live change\nreplaces a running component, Archie starts and checks the new one before\nswitching. If the check fails, the old one keeps running.\n\n## Bootstrap, migration, and recovery\n\nBootstrap contains only the database path, listen addresses, credentials, TLS,\nand encryption-key references. State Store and control-plane RPCs use the same\nendpoint.\n\nWhen upgraded, existing instances are migrated to the database. Existing\ndatabase values are not overwritten, with one exception: a resource whose\ndocument the build ships is refreshed when the seed itself wrote the stored\ncopy, because that copy is a previous release's document rather than an\noperator's. After migration, settings in TOML are ignored and cannot block\nState Store startup.\n\nThe Web UI remains available when `archied` or Gateway is down. It shows failed\nchanges and can correct or roll them back.\n\nOffline commands can back up, restore, validate, and roll back the database.\nRecovery does not require `archied` or the Web UI. The database stores secret\nreferences, not secret values.\n\n## Workflows\n\nUsers create and edit workflows in the Web UI. Workflows are stored as versioned\nYAML. Built-in workflows can be overridden and restored.\n\nA workflow is a list of steps. Each step has a known type and its own settings.\nA step type is part of the build's step vocabulary, contributed as a provider at\nthe composition root; interpreted plugins contribute none\n(`docs/prds/workflow-step-vocabulary.md`). Each run keeps the workflow version it\nstarted with.\n\nYaegi, `.archie/stages/*.go`, and `.archie/gate.go` will be removed. Their useful\nbehaviour must first be available as normal workflow steps.\n\n## First implementation\n\nThe first implementation makes task limits editable: maximum model steps,\nmaximum runtime, and maximum consecutive gate failures.\n\nIt includes:\n\n- validated task-limit fields;\n- version history and audit records;\n- the four API operations;\n- authenticated Web UI and messaging access;\n- live updates and apply status;\n- a PiLine truncated
},
{
"url": "/dev/prds/self-update-transaction/",
@@ -562,7 +562,7 @@
"url": "/dev/prds/workflow-call-outputs/",
"title": "Declared outputs of a workflow run",
"section": "PRDs",
"body": "# Declared outputs of a workflow run\n\n**Status:** Draft\n**Date:** 2026-09-30\n**Bead:** `archie-core-t2db.43`\n**Authority:** `docs/prds/event-automation.md` for what a declared output is,\n`docs/prds/workflow-calls.md` for the call and its grants,\n`docs/prds/state-store-contract.md` for the State Store contract the wire\nchange must satisfy.\n**Supersedes:** `docs/prds/workflow-calls.md`, \"Waiting on the callee\": the\ncallee's transition detail is the human summary, not the stand-in for\nstructured outputs.\n\nA workflow declares named, typed outputs. A run writes each one once. A\n`wait: true` caller receives them, and may publish one as its own.\n\n## Where outputs are declared\n\n`outputs:` on the workflow definition, beside `inputs:`, decoded into\n`task.WorkflowInterface.Outputs map[string]OutputSpec` where\n`OutputSpec{Type, Required}`:\n\n```yaml\nid: firewall-contain\noutputs:\n contained: { type: bool, required: true }\n summary: { type: string }\n```\n\nNames and types are the input grammar: the same identifier shape, the same\ntype set, and the same check in `WorkflowInterface.Validate`, which both\n`ParseDefinition` and `ParseWorkflowInterface` call.\n\n**Forced by:** `ParseDefinition` decodes with `KnownFields(true)`, so a key no\ndecoded struct declares is refused — the declaration must live on\n`WorkflowInterface` for an author to write it at all, and that is where\n`docs/development/workflows.md` puts every field a process without the engine\nmust read. The type set is the shared value vocabulary: `TypeAccepts` and\n`ValueType` already check a mapped parameter and an input with it. `required`\ndefaults to false, and the generated YAML for a builtin that declares no\noutputs is unchanged, because a definition's digest pins a run.\n\n**Rejected:** outputs per step, or per call step, because a workflow returns a\nresult as itself rather than as one of its stages. Also `task.Definition`\ngaining outputs: no reader needs them — a binding offers inputs, and a call's\noutputs assignment is checked against the callee's YAML in the same\ncollection — and a declared field with no reader parses and does nothing.\n\n## How a run writes one\n\nEach declared output is offered to the run's agent stages as a capture tool\nnamed after the output, whose arguments are `{\"value\": ...}`; the accepted\ncall's `value` member is the output's value. The engine holds the accepted\nvalues on the task context and writes them into its own task row in the\n`Store.Update` call the run already makes. A deterministic stage may set the\nsame values in Go, and the run's finish check validates them identically.\n\n**Forced by:** a run's only structured results are capture calls\n(`agentexec.CaptureTool`, `Result.Captures`), and the workflow interface\nalready declares when a stage needs them (`task.WorkflowNeeds.Captures`).\nDeriving the tools from the declaration means an output cannot be written\nunder a name nobody declared, and it keeps one route to structured agent\noutput rather than adding a second beside the capture path. The row write is\nthe run's existing write: the task row is where a run's own state lives, and\nthe container already holds `Update` on its own row.\n\nThe tool is offered with `MaxCalls: 1`, and a stage is not offered an output\nthe run has written, so a run writes each output once: a repeat within a stage\nis answered with the tool's rejection and the agent carries on, and a later\nstage cannot overwrite a value. A stage's own capture tool may not share a\nname with a declared output, because the value would otherwise be ambiguous.\n\n**Rejected:** a `workflow.output` step type, because assigning the run's\nresult from a step needs a grammar for \"the previous stage's result\" that does\nnot exist, and a step that reaches the agent would be the second path\n`docs/development/workflows.md` forbids. Also a write RPC of its own: the\ncontainer's `Update` already reaches its own row, and a second write verb for\none column is a second thing to keep in step.\n\n## How a caller receives one\n\nOn a successful `wait: true` call the caller's stage holds the callee's\noutputs as a `map[string]any`. A call step's `outputs:` setting publishes one\nas one of the caller's own outputs:\n\n```yaml\nsteps:\n - type: workflow.call\n workflow: firewall-contain\n wait: true\n inputs: { src_ip: inputs.src_ip }\n outputs: { contained: outputs.contained }\n```\n\nThe key names the callee's declared name; the value names one of the caller's\nown with the prefix the call's `inputs:` already uses. A published value joins\nthe caller's output set for its run, so a caller returns its callee's result to\nits own caller, and a run may publish a callee's output and write its own\nothers.\n\nLifetime is the attempt. The run starts with an empty output set, every later\nstage of that attempt sees the values, and the finish write replaces the row's\nset wholesale — so nothing an attemLine truncated
"body": "# Declared outputs of a workflow run\n\n**Status:** Draft\n**Date:** 2026-09-30\n**Bead:** `archie-core-t2db.43`\n**Authority:** `docs/prds/event-automation.md` for what a declared output is,\n`docs/prds/workflow-calls.md` for the call and its grants,\n`docs/prds/state-store-contract.md` for the State Store contract the wire\nchange must satisfy.\n**Supersedes:** `docs/prds/workflow-calls.md`, \"Waiting on the callee\": the\ncallee's transition detail is the human summary, not the stand-in for\nstructured outputs.\n\nA workflow declares named, typed outputs. A run writes each one once. A\n`wait: true` caller receives them, and may publish one as its own.\n\n## Where outputs are declared\n\n`outputs:` on the workflow definition, beside `inputs:`, decoded into\n`task.WorkflowInterface.Outputs map[string]OutputSpec` where\n`OutputSpec{Type, Required}`:\n\n```yaml\nid: firewall-contain\noutputs:\n contained: { type: bool, required: true }\n summary: { type: string }\n```\n\nNames and types are the input grammar: the same identifier shape, the same\ntype set, and the same check in `WorkflowInterface.Validate`, which both\n`ParseDefinition` and `ParseWorkflowInterface` call.\n\n**Forced by:** `ParseDefinition` decodes with `KnownFields(true)`, so a key no\ndecoded struct declares is refused — the declaration must live on\n`WorkflowInterface` for an author to write it at all, and that is where\n`docs/development/workflows.md` puts every field a process without the engine\nmust read. The type set is the shared value vocabulary: `TypeAccepts` and\n`ValueType` already check a mapped parameter and an input with it. `required`\ndefaults to false, and the generated YAML for a builtin that declares no\noutputs is unchanged, because a definition's digest pins a run.\n\n**Rejected:** outputs per step, or per call step, because a workflow returns a\nresult as itself rather than as one of its stages. Also `task.Definition`\ngaining outputs: no reader needs them — a binding offers inputs, and a call's\noutputs assignment is checked against the callee's YAML in the same\ncollection — and a declared field with no reader parses and does nothing.\n\n## How a run writes one\n\nEach declared output is offered to the run's agent stages as a capture tool\nnamed after the output, whose arguments are `{\"value\": ...}`; the accepted\ncall's `value` member is the output's value. The engine holds the accepted\nvalues on the task context and writes them into its own task row in the\n`Store.Update` call the run already makes. A deterministic stage may set the\nsame values in Go, and the run's finish check validates them identically.\n\n**Forced by:** a run's only structured results are capture calls\n(`agentexec.CaptureTool`, `Result.Captures`), and the workflow interface\nalready declares when a stage needs them (`task.WorkflowNeeds.Captures`).\n`WorkflowInterface.Needs` reports `captures` for a non-empty `Outputs`, so\ndeclaring an output alone requires the capability and a Kit profile that\nserves no capture tools is refused before dispatch.\nDeriving the tools from the declaration means an output cannot be written\nunder a name nobody declared, and it keeps one route to structured agent\noutput rather than adding a second beside the capture path. The row write is\nthe run's existing write: the task row is where a run's own state lives, and\nthe container already holds `Update` on its own row.\n\nThe tool is offered with `MaxCalls: 1`, and a stage is not offered an output\nthe run has written, so a run writes each output once: a repeat within a stage\nis answered with the tool's rejection and the agent carries on, and a later\nstage cannot overwrite a value. A stage's own capture tool may not share a\nname with a declared output, because the value would otherwise be ambiguous.\n\n**Rejected:** a `workflow.output` step type, because assigning the run's\nresult from a step needs a grammar for \"the previous stage's result\" that does\nnot exist, and a step that reaches the agent would be the second path\n`docs/development/workflows.md` forbids. Also a write RPC of its own: the\ncontainer's `Update` already reaches its own row, and a second write verb for\none column is a second thing to keep in step.\n\n## How a caller receives one\n\nOn a successful `wait: true` call the caller's stage holds the callee's\noutputs as a `map[string]any`. A call step's `outputs:` setting publishes one\nas one of the caller's own outputs:\n\n```yaml\nsteps:\n - type: workflow.call\n workflow: firewall-contain\n wait: true\n inputs: { src_ip: inputs.src_ip }\n outputs: { contained: outputs.contained }\n```\n\nThe key names the callee's declared name; the value names one of the caller's\nown with the prefix the call's `inputs:` already uses. A published value joins\nthe caller's output set for its run, so a caller returns its callee's result to\nits own caller, and a run may publish a callee's output and write its own\notheLine truncated
},
{
"url": "/dev/prds/workflow-calls/",
@@ -574,7 +574,7 @@
"url": "/dev/prds/workflow-step-vocabulary/",
"title": "Workflow step vocabulary: what a step type is and how binaries agree on it",
"section": "PRDs",
"body": "# Workflow step vocabulary: what a step type is and how binaries agree on it\n\n**Status:** Draft\n**Date:** 2026-09-29\n**Beads:** `archie-core-m7oc` (agreement between separately deployed binaries), `archie-core-ze7y` (the deleted `.archie/stages/*.go` equivalent)\n**Extends:** `docs/architecture/plugins-and-extensions.md` (plugin engine rule), `docs/prds/event-automation.md` (the workflow model step types serve)\n\n**Authority:** `docs/architecture/plugins-and-extensions.md` — the workflow step vocabulary is a Workflow capability family, and its extension point follows the plugin engine rule.\n\n## Decision\n\nA workflow step type is a named Go factory: the word a YAML workflow writes in\na step's `type` field, plus the factory that builds its stage. The name is a\nstable identifier; the identity also declares what plugin family claims it,\nso a contribution that shadows a shipped stage is refused instead of\noverriding it.\n\nThe Workflow domain owns the typed family contract — `StepTypeProvider` for a\ncontribution and `Manager` as the owning registry. A provider contributes a\nwhole set of step types under its own stable-identifier name; the manager\nvalidates the entire contribution before applying any of it (named provider,\nwell-formed step-type names, non-nil factories, no step type claimed twice,\nno claim on a name another provider holds) and applies it atomically. The\ngeneric `plugin.Plugin` contract stays metadata-only.\n\nThe vocabulary is a compiled-in, closed set. `internal/infrastructure/workflowsteps`\nis the one place a step type is bundled into any binary that resolves one: the\nshipped workflow stages, the repository-hook replacements (`workflow.diff-rules`)\nand the repository-free event steps each arrive as one provider, and adding a\nstep type means adding a provider to that set. Interpreted plugin sources\ncannot contribute step types: the Yaegi plugin host satisfies only the\nmetadata contract, and no bridge to `StepFactory` exists.\n\nEvery binary that resolves a step type builds its manager from that same\nprovider set at its composition root, before its first resolution, and\ninjects the manager everywhere it resolves definitions. Agreement is per\nbuild: the State Store, the daemon's workflow-definitions client and the agent\nworker built from the same source register identical vocabularies by\nconstruction; binaries built from different sources can disagree, and only a\nmatching deploy fixes that.\n\nAn unknown step type is refused twice, and never degraded. The validator\n`ParseDefinition` refuses a definition naming an unregistered type at save\ntime, on the validating side the composition root built (the State Store's\nworkflow-definitions resource validates there before anything is stored). The\ncompiler refuses the same definition at run, so a binary handed a definition\nwritten against a vocabulary it does not have fails that run loudly instead of\nsilently skipping steps. Both checks live in the Workflow domain against the\ninjected manager, which is why no binary can opt out of the check by\nsubstituting its own registry.\n\n## Relationship between the questions\n\n`archie-core-ze7y` is the same question seen from the deleted\nrepository-authored stages: those `.archie/stages/*.go` files held arbitrary\nstages, and their gate half landed as `workflow.diff-rules` settings rather\nthan interpreted Go. Their stage half has no replacement and does not get one\nthat approximates it: no interpreted-Go stage type is built, and no command\nstep type is created by this decision to stand in for them. The refusal of\n`.archie/stages/*.go` stays until the maintainer's trust decision\n(`archie-core-ndm8`) says otherwise.\n\nIf that decision ever allows a command step type, it arrives as a provider in\nthe compiled-in set under the rules above — a new contributor of exactly one\nstep type at a named provider. It is not a plugin contribution mechanism, and\nit does not change any answer in this document.\n\n## Rejected alternatives\n\n- **A vocabulary version or hash on the wire.** The executing side already\n refuses an unknown step type at its first compile, which is the mismatch\n detector, and it fails closed. A wire field would only narrow the\n deploy-skew window that a matching deploy closes, at the cost of a new\n contract field.\n- **A Yaegi-plugin-to-`StepFactory` bridge.** It would run repository-supplied\n interpreted code inside the daemon's process with daemon privileges. The\n trust boundary invariant requires repository-authored code to run in the\n task container, not in an interpreter embeddable into any binary.\n- **A package-level process-global registrar.** Global mutable registry state\n is unreadable at the composition root, collides with itself when one\n process builds the vocabulary twice, and is what made the first attempt at\n this question worse instead of better.\n- **Degrading on an unknown step type.** A run that skipsLine truncated
"body": "# Workflow step vocabulary: what a step type is and how binaries agree on it\n\n**Status:** Draft\n**Date:** 2026-09-29\n**Beads:** `archie-core-m7oc` (agreement between separately deployed binaries), `archie-core-ze7y` (the deleted `.archie/stages/*.go` equivalent)\n**Extends:** `docs/architecture/plugins-and-extensions.md` (plugin engine rule), `docs/prds/event-automation.md` (the workflow model step types serve)\n\n**Authority:** `docs/architecture/plugins-and-extensions.md` — the workflow step vocabulary is a Workflow capability family, and its extension point follows the plugin engine rule.\n\n## Decision\n\nA workflow step type is a named Go factory: the word a YAML workflow writes in\na step's `type` field, plus the factory that builds its stage. The name is a\nstable identifier; the identity also declares what plugin family claims it,\nso a contribution that shadows a shipped stage is refused instead of\noverriding it.\n\nThe Workflow domain owns the typed family contract — `StepTypeProvider` for a\ncontribution and `Manager` as the owning registry. A provider contributes a\nwhole set of step types under its own stable-identifier name; the manager\nvalidates the entire contribution before applying any of it (named provider,\nwell-formed step-type names, non-nil factories, no step type claimed twice,\nno claim on a name another provider holds) and applies it atomically. The\ngeneric `plugin.Plugin` contract stays metadata-only.\n\nThe vocabulary is a compiled-in, closed set. `internal/infrastructure/workflowsteps`\nis the one place a step type is bundled into any binary that resolves one: the\nshipped workflow stages, the repository-hook replacements (`gate.diff-rules`)\nand the repository-free event steps each arrive as one provider, and adding a\nstep type means adding a provider to that set. Interpreted plugin sources\ncannot contribute step types: the Yaegi plugin host satisfies only the\nmetadata contract, and no bridge to `StepFactory` exists.\n\nEvery binary that resolves a step type builds its manager from that same\nprovider set at its composition root, before its first resolution, and\ninjects the manager everywhere it resolves definitions. Agreement is per\nbuild: the State Store, the daemon's workflow-definitions client and the agent\nworker built from the same source register identical vocabularies by\nconstruction; binaries built from different sources can disagree, and only a\nmatching deploy fixes that.\n\nAn unknown step type is refused twice, and never degraded. The validator\n`ParseDefinition` refuses a definition naming an unregistered type at save\ntime, on the validating side the composition root built (the State Store's\nworkflow-definitions resource validates there before anything is stored). The\ncompiler refuses the same definition at run, so a binary handed a definition\nwritten against a vocabulary it does not have fails that run loudly instead of\nsilently skipping steps. Both checks live in the Workflow domain against the\ninjected manager, which is why no binary can opt out of the check by\nsubstituting its own registry.\n\n## Relationship between the questions\n\n`archie-core-ze7y` is the same question seen from the deleted\nrepository-authored stages: those `.archie/stages/*.go` files held arbitrary\nstages, and their gate half landed as `gate.diff-rules` settings rather\nthan interpreted Go. Their stage half has no replacement and does not get one\nthat approximates it: no interpreted-Go stage type is built, and no command\nstep type is created by this decision to stand in for them. The refusal of\n`.archie/stages/*.go` stays until the maintainer's trust decision\n(`archie-core-ndm8`) says otherwise.\n\nIf that decision ever allows a command step type, it arrives as a provider in\nthe compiled-in set under the rules above — a new contributor of exactly one\nstep type at a named provider. It is not a plugin contribution mechanism, and\nit does not change any answer in this document.\n\n## Rejected alternatives\n\n- **A vocabulary version or hash on the wire.** The executing side already\n refuses an unknown step type at its first compile, which is the mismatch\n detector, and it fails closed. A wire field would only narrow the\n deploy-skew window that a matching deploy closes, at the cost of a new\n contract field.\n- **A Yaegi-plugin-to-`StepFactory` bridge.** It would run repository-supplied\n interpreted code inside the daemon's process with daemon privileges. The\n trust boundary invariant requires repository-authored code to run in the\n task container, not in an interpreter embeddable into any binary.\n- **A package-level process-global registrar.** Global mutable registry state\n is unreadable at the composition root, collides with itself when one\n process builds the vocabulary twice, and is what made the first attempt at\n this question worse instead of better.\n- **Degrading on an unknown step type.** A run that skips steps iLine truncated
}
],
"nav": [
+9
View File
@@ -43,6 +43,15 @@ and a decision record for when it is deliberately not enough:
draft state, copy conditions, filters and payload building live in plain
`.ts` files beside the components and are tested in `ui/test/`. Components
stay thin, so there is normally nothing in an SFC that is archie logic.
- **A dashboard rule that mirrors a server rule is pinned by a shared
fixture, not two independent tests.** `ui/src/settings/validation.ts`
hand-mirrors three control-plane validators, and the model-role rule had
already drifted because nothing tied the two sides together. Each case now
lives once in `internal/app/controlplane/testdata/dashboard-validators.json`:
a Go test in that package runs the real server validators over it, and
`ui/test/control-plane-validator-parity.test.ts` runs the real dashboard
validators over the same cases. Changing one rule alone fails the other
side's test. Add a case to the fixture, never to one side only.
- **Installed primitives are upstream-tested, not re-tested here.**
`reka-ui` (2.10.5 here) maintains colocated tests per component in its own
repo (Vitest/jsdom + axe: Switch, TagsInput, NumberField, Select, the
+17
View File
@@ -28,6 +28,19 @@ vocabulary it registers, so both must see the same step types.
6. Add a parse test (settings accepted and refused) and a run test with fake
`Trees` and agent runner.
## Retiring a step type
A step word cannot simply be deleted. A stored or pinned definition is
validated as a whole collection, so one unknown word fails EVERY workflow's
definition to decode and the workflow-definitions resource becomes unreadable.
Retire a stage by removing it from its workflow's `Stages` and adding it to
`retiredSteps` in `definition.go`, which keeps the word resolving as an inert
stage. The retired stage must do nothing: the behaviour it used to perform now
belongs to whatever layer took it over (the remediate workflow's in-container
resume moved to daemon worktree preparation, `archie-core-866m`; the workflow
still binds the daemon-prepared directory and the PR branch in a `prepare`
stage before any agent work).
## Changing the definition format
- The fields a caller needs without the engine (inputs, repository, profile)
@@ -37,6 +50,10 @@ vocabulary it registers, so both must see the same step types.
- The YAML decoder rejects unknown keys. Existing stored definitions must keep
parsing, so a new key is optional with a default that preserves today's
behaviour.
- A declared `outputs:` entry is written through a capture tool, so it forces
the captures need: `task.WorkflowInterface.Needs` resolves that from the
declaration rather than the author also writing `needs.captures`. A Kit
profile whose harness serves no capture tools is refused before dispatch.
- Examples in `examples/workflows/` are parsed against the registered
vocabulary by `TestExampleWorkflowsParse`. Add one for a new capability.
+2
View File
@@ -60,6 +60,7 @@
"/news/archie/1.44.0/": "/news/1.44.0/",
"/news/archie/1.44.1/": "/news/1.44.1/",
"/news/archie/1.45.0/": "/news/1.45.0/",
"/news/archie/1.46.0/": "/news/1.46.0/",
"/news/archie/1.5.0/": "/news/1.5.0/",
"/news/archie/1.6.0/": "/news/1.6.0/",
"/news/archie/1.7.0/": "/news/1.7.0/",
@@ -137,6 +138,7 @@
"/news/archied/1.44.0/": "/news/1.44.0/",
"/news/archied/1.44.1/": "/news/1.44.1/",
"/news/archied/1.45.0/": "/news/1.45.0/",
"/news/archied/1.46.0/": "/news/1.46.0/",
"/news/archied/1.5.0/": "/news/1.5.0/",
"/news/archied/1.6.0/": "/news/1.6.0/",
"/news/archied/1.7.0/": "/news/1.7.0/",
+5
View File
@@ -1,4 +1,9 @@
[
{
"version": "1.46.0",
"date": "2026-10-02",
"body": "### archied\n\n- feat(workflow): declared workflow outputs for wait:true callers\n- feat(prreview): the operator-approval gate is PRReview's own stage-list entry\n- feat(staterpc): carry the review gate and re-review rounds, with one guarded response write\n- feat(prreview): the gate records the review it holds and posts the operator's answer\n- feat(taskactions): one approval path answers the review gate from every surface\n- feat(webui): the dashboard answers the review gate, including a re-review's instructions\n- feat(daemon): resolve the producing identity's org before publishing work\n- feat(webui): block invalid settings and show a field's former value\n- fix(workflow): derive a declared output's captures need for profile selection\n- fix(storepkg): validate every declared authority grant at the package descriptor\n- feat(archied): refresh the model catalog at runtime\n- feat(mcp): answer server-initiated sampling from the chat model\n- feat(container): send registry auth on image pull\n- feat(messaging): retract ephemeral replies after their TTL\n- feat(messaging): transcribe Telegram voice notes\n- fix(egress): take a run's credential kind from its binding, not the Kit\n- feat(webui): a live-apply setting reads as applied, never as a restart\n- feat(webui): the dashboard reads harness bindings and opens the setup terminal\n- feat(gateway): list the harness page in the dashboard page registry\n- feat(webui): the harness settings page and its setup terminal\n- fix(workflow): park environment gate failures without dispatching a builder\n- feat(messaging): adapter-facing clarify, picker and approval contract\n- feat(gateway): gate admin slash commands by sender policy\n- feat(plugins): load plugin and secret-engine files without a restart\n- fix(workflow): deliver the feasibility PRD to the forge issue\n- fix(messaging): report the daemon version this process was built from\n- fix(messaging): carry an inbound attachment's bytes to the gateway turn\n- fix(daemon): resume a retry whose worktree is gone\n- fix(archied): omit the max-token bound on reasoning-class models\n- fix(logging): file a task log record under its own event time\n- fix(control-plane): unify validation for the domain-managed kinds\n- feat(webui): let the operator choose the retry worktree mode\n- feat(agent): seed a starter SOUL and upgrade only untouched templates\n- feat(controlplane): add the writable review-settings resource\n- fix(messaging): transcribe voice notes on the model-owning side of the boundary\n- fix(messaging): gate the boundary on config reads, not only package links\n- feat(messaging): restart only the channel whose settings changed\n- feat(container): apply container-runtime-policies live\n- feat(tools): reconcile MCP servers and tool settings live\n- feat(messaging): add a per-turn interactive carrier for clarify and picker\n- feat(gateway): offer ask_user when the turn's channel can carry a question\n- feat(telegram): carry clarify and picker through the text fallback\n- fix(webui): pin the dashboard validators to the control plane's\n- refactor(agent): drop the SOUL upgrade branch with no shipped template\n- feat(agentworker): answer MCP sampling from the task's model\n- fix(workflow): bind the remediate worktree before the builder runs\n- fix(config): carry max_retries into the worker TaskConfig snapshot\n- fix(agentworker): supply the guardrail engine to every task run\n\n### archie-agent\n\n- feat(workflow): declared workflow outputs for wait:true callers\n- feat(prreview): the operator-approval gate is PRReview's own stage-list entry\n- feat(staterpc): carry the review gate and re-review rounds, with one guarded response write\n- feat(prreview): the gate records the review it holds and posts the operator's answer\n- feat(taskactions): one approval path answers the review gate from every surface\n- fix(workflow): derive a declared output's captures need for profile selection\n- fix(storepkg): validate every declared authority grant at the package descriptor\n- feat(mcp): answer server-initiated sampling from the chat model\n- feat(container): send registry auth on image pull\n- feat(messaging): retract ephemeral replies after their TTL\n- feat(messaging): transcribe Telegram voice notes\n- fix(egress): take a run's credential kind from its binding, not the Kit\n- fix(workflow): park environment gate failures without dispatching a builder\n- feat(messaging): adapter-facing clarify, picker and approval contract\n- feat(plugins): load plugin and secret-engine files without a restart\n- fix(workflow): deliver the feasibility PRD to the forge issue\n- fix(messaging): carry an inbound attachment's bytes to the gateway turn\n- fix(daemon): resume a retry whose worktree is gone\n- fix(logging): file a task log record under its own event time\n- fix(control-plane): unify validation for the domain-managed kinds\n- feat(webui): let the operator choose the retry worktree mode\n- fix(messaging): transcribe Line truncated
},
{
"version": "1.45.0",
"date": "2026-09-30",
+13 -5
View File
@@ -56,11 +56,18 @@ messages by method name (the same switch that already routes
`internal/app/archied/main.go` that routes the sampling request through the
same chat-model call path `chatGenerateOptions` already uses, so a
server-initiated sample gets the daemon's configured model, not a second
provider client.
provider client. The agent worker hosts the same configured MCP servers on the
task path and runs its own model runtime, so `internal/app/agentworker` wires
the same seam from the task's builder model: a server that samples succeeds
inside a chat turn and inside a task, never one and not the other. The message
mapping and the token bound are shared (`internal/agentexec`, next to the
runtime both paths call models through) so both paths bound a sample
identically.
**Reuse check.** The chat model invocation path already exists
(`chatGenerateOptions`, `main.go`); this reuses it rather than adding a
parallel model-calling path for MCP sampling.
(`chatGenerateOptions`, `main.go`) and the task model path already exists
(`agentexec.NewRuntime`, the runtime an agent stage runs on); this reuses both
rather than adding a parallel model-calling path for MCP sampling.
**Failure mapping.**
@@ -170,7 +177,8 @@ preview callbacks fire in order before the next call starts. `task check`.
| MCP client call serialisation | `internal/tools/mcp/client.go` | change (`#177`) |
| MCP server config | `internal/config/config.go` | new fields: `ParallelToolCalls`, `Sandboxed` |
| Sampling dispatch | `internal/tools/mcp/client.go` | new (`#151`) |
| Sampling handler wiring | `internal/app/archied/main.go` | new (`#151`), reuses `chatGenerateOptions` |
| Sampling handler wiring (chat path) | `internal/app/archied/main.go` | new (`#151`), reuses `chatGenerateOptions` |
| Sampling handler wiring (task path) | `internal/app/agentworker/mcp_sampling.go`, `mcp_providers.go` | new (`#151`), answers from the task's builder model |
| Sandboxed MCP launch | MCP client constructor + `internal/container/pool.go` | new (`#178`/`#179`) |
| Concurrent tool dispatch | dispatch entry point (closed `#183`) | none — already handles this |
| Sequential tool dispatch + previews | same dispatch entry point | new (`#182`) |
@@ -181,7 +189,7 @@ preview callbacks fire in order before the next call starts. `task check`.
| sub-feature | issue | implementer scope | suggested council lenses | why |
| ------------------------------------ | -------------- | ---------------------------------------------------------------------------------- | ----------------------------------- | ----------------------------------------------------------------------------------------------- |
| Server-initiated sampling | `#151` | `client.go` dispatch branch, `main.go` handler wiring | `lens-contract` | new JSON-RPC method surface crossing the MCP wire contract |
| Server-initiated sampling | `#151` | `client.go` dispatch branch, chat-path and task-path handler wiring | `lens-contract` | new JSON-RPC method surface crossing the MCP wire contract |
| Parallel-tool-calls flag | `#177` | config field, conditional mutex, constructor wiring | `lens-boundary` | per-server behaviour change must not leak into other servers' serialisation |
| Sandboxed MCP + Firecracker deferral | `#178`, `#179` | config field, constructor routing through `ContainerPool`, pool entrypoint variant | `lens-operator`, `lens-deletionist` | operator: sandbox-down failure path; deletionist: keeps `#179` from becoming unused scaffolding |
| Sequential dispatch with previews | `#182` | `DispatchMode` enum, sequential branch, preview reuse of streaming seam | `lens-maintainer` | two coexisting dispatch modes need a reader to tell which applies without re-deriving both |
+100 -54
View File
@@ -1,6 +1,7 @@
# Plugin settings reconcile live
**Status:** Draft
**Revision:** 2
**Tracking:** GitHub [#1154](https://github.com/samcharles93/archie-core/issues/1154), epic [#998](https://github.com/samcharles93/archie-core/issues/998)
**Authority:** `docs/prds/runtime-control-plane.md` decides each feature owns how
its changes take effect; `docs/architecture/plugins-and-extensions.md` owns the
@@ -9,34 +10,38 @@ reports go.
## Outcome
An operator drops a plugin, module or secret-engine file into the configured
An operator drops a plugin, module or secret-engine file into a configured
directory while archied runs, and the daemon loads it without a restart. An
edit to an installed file replaces what is running. Deleting an installed file
never stops running code; apply status says so. A stored change to
`plugin-settings` applies live for additions instead of requiring a restart.
edit to an installed file replaces what is running. A delete never stops running
code; apply status says so. A stored `plugin-settings` change applies live for
`plugin_dir`, `module_dir` and `secret_engine_dir`; a change to `skills_dir` is
refused.
## Triggers
`plugin-settings` joins the kinds archied watches (`PluginSettingsKind` moves
to the live list in `internal/app/archied/control_plane.go`). After the daemon
layers a new version it reconciles the three directories the document names.
A second trigger needs no document change: a poller reconciles on an interval,
so a file dropped into a directory loads on its own. The poll interval is the
apply-status restamp constant; no filesystem-watch dependency is added. The
poller reads the current document's directories from the running config each
tick, so a changed `plugin_dir`, `module_dir` or `secret_engine_dir` retargets
reconciliation without a restart.
`PluginSettingsKind` joins `runtimeResourceKinds` in
`internal/app/archied/control_plane.go`, and its descriptor `ApplyMode` becomes
`live` in `internal/app/controlplane/operational_settings.go`.
A poller reconciles on the apply-status restamp interval, so a dropped file
loads without a filesystem watch. Each tick re-runs both halves the boot path
runs, against the directories in the running config: the directory loads
(`secret.Registry.LoadDir`, `ModuleRegistry.Register`, `plugin.LoadDir`) and the
layering (`boot.runtimeConfig`), which re-resolves provider credentials. A
provider disabled at boot for an unresolvable engine is re-resolved on the first
tick that finds it; a changed directory retargets the next tick without a
restart.
## Rules
- Adding and changing load; removing never unloads. Yaegi interpreters are not
reclaimed while a caller still exists and there is no unload path; the
running set can only grow.
- Each file is loaded through a fresh interpreter (the boot pattern). A changed
file therefore loads cleanly, and replacement switches consumers to the new
interpretation; files are content-hashed so only new and changed files load.
reclaimed while a caller still exists and there is no unload path; the running
set can only grow. A removal stays outstanding in the apply-status error until
the process restarts.
- Each file is loaded through a fresh interpreter (the boot pattern) and
content-hashed, so only new and changed files load.
- A file that fails to load is skipped and reported; already-running code keeps
running. This is the degrade-and-skip convention `plugin.LoadDir` and
running, the degrade-and-skip convention `plugin.LoadDir` and
`Registry.LoadDir` already apply at boot. Boot keeps its reject-at-load rule
for the module directory; a broken file at runtime skips without aborting.
- A directory that does not exist reconciles to an empty set.
@@ -44,45 +49,86 @@ reconciliation without a restart.
## The three registries
- **Secret engines** (`internal/secret`). Reconciliation re-runs
`Registry.LoadDir` against the live directory. `Register` replaces by name,
so an edited engine takes effect for the next resolution. A deleted engine
stays registered: existing references keep resolving through it until
restart, and the record says so. Engines are stateless resolvers, so a swap
needs no lifecycle step.
- **Modules** (`internal/domain/eda/module`): reconciliation re-runs
`ModuleRegistry.Register` for the kind's file. The invoker maps
last-write-wins, and playbook dispatch resolves through the registry at
invoke time, so the swap reaches a running playbook store unchanged. A file
deleted for a kind leaves the old invoker in place, running; reported.
- **Plugins** (`internal/plugin`): `Host` grows running-state operations beside
`Register`, `Start` and `Stop`. One adds a single module while the host is
running: it is inserted in dependency order, started alone, and on failure
rolled back alone, leaving every previously started module running. The
other replaces a module whose manifest id matches: stop it, start the new
one, and restart the old one when the new start fails. Replacing a module
that another depends on is refused, not swapped. These operations obey the
plugin engine rule: the host still invokes start and stop itself, isolates
the failure to the one module, and keeps a rollback. A deleted file makes no
host call: the module runs on and is reported.
`Registry.LoadDir` against the live directory. `Register` replaces by name, so
an edited engine takes effect for the next resolution and a dropped engine
resolves on the next tick. A deleted engine stays registered: existing
references keep resolving through it until restart. Engines are stateless
resolvers, so a swap needs no lifecycle step.
- **Modules** (`internal/domain/eda/module`). Reconciliation re-runs
`ModuleRegistry.Register` for the kind's file. Playbook dispatch resolves
through the registry at invoke time, so the swap reaches a running playbook
store unchanged; a deleted file leaves the old invoker in place, running.
- **Plugins** (`internal/plugin`). `Host` grows running-state operations beside
`Register`, `Start` and `Stop`, and `Register` keeps its rule that
registration is allowed only before the first `Start`, so these are new host
methods rather than a relaxed `Register`. One adds a single module while the
host is running: inserted in dependency order, started alone, and rolled back
alone on failure, leaving every previously started module running. The other
replaces a module whose manifest id matches: stop it, start the new one, and
restart the old one when the new start fails; replacing a module another
running module depends on is refused. A module's `Start` is its only check
(`plugin.Module` has no probe, and one manifest id cannot be registered
twice), so a replacement checks by starting and keeps a rollback rather than
switching first. These operations obey the plugin engine rule: the host
invokes start and stop itself and isolates the failure to one module. A
deleted file makes no host call: the module runs on.
## `skills_dir`
`skills_dir` is not part of this change. The daemon refuses a candidate
`plugin-settings` document whose `skills_dir` differs from the value it is
running, reporting that the field applies on restart, until a skill
reconciliation exists to carry it. The refusal is the existing rejected-edit
apply-status shape: the store keeps the version, the document is not applied,
`skills_dir` is not part of this change. The daemon refuses a `plugin-settings`
document whose `skills_dir` differs from the value in force. The check is a
pre-layer comparison on the live path (`boot.applyRuntimeResourceUpdate` in
`internal/app/archied/control_plane.go`), not an addition to
`validatePluginSettings` in `internal/app/controlplane/operational_settings.go`,
which stays value-agnostic per the #1143 decision recorded there: a
change-immutability rule needs the running config, and #1143's validates-nothing
still holds for values. The write lands, the store keeps the version, the daemon
publishes nothing, the refusal is reported against that stored version, and
last-known-good keeps running.
The layering assigns all four directories from one document, so a refused
`skills_dir` refuses the whole document and the other kinds still layer. That is
acceptable: `skills_dir` is the one field with no live consumer, a
whole-document refusal is the honest signal that none of it took effect, and an
operator changing `skills_dir` alongside another directory writes the two
changes separately.
## Apply status
One record: `archied` on `plugin-settings`. The reconciliation outcome is
reported against the version the process is running. The record's error field
carries the outstanding reconciliation problem: a deleted file still running, a
file that failed to load, or the directory failing to read. A reconciliation
with nothing outstanding clears the error. The version stays the version the
process is running; an add that succeeded is live even while a removal is
outstanding, and the error says so rather than showing the kind as
restart-required. archie-gateway keeps boot-time loading; its record simply
stays on the version it last applied until it restarts, which the per-process
apply-status page already reads correctly.
One record per process: `archied` on `plugin-settings`, reported against the
version the process is running. The record's error field carries the outstanding
problem: a deleted file still running, a file that failed to load, the directory
failing to read, or a refused replace of a module another running module depends
on. A clean reconciliation clears the error. An add that succeeded is live even
while a removal is outstanding, and the error says so rather than showing the
kind as restart-required.
`ApplyMode` becomes `live`, so `RestartPendingBanner` no longer names
`plugin-settings`. The gateway's lag is shown instead by the per-process
applied-version rows on the Plugins page, which read the state
`applyStatusForKind` derives from `applied_version < storedVersion` in
`ui/src/stores/control-plane.ts`; that derivation is independent of
`ApplyMode`.
`archie-gateway` layers the `plugin-settings` document at boot through
`boot.loadRuntimeConfig` and reports it, but never calls `boot.loadPlugins` or
`boot.loadModules` and never reads the four directories, so it loads no plugin,
module or secret engine. With no live watch its row lags the stored version
until restart; that lag is a reporting lag, not a behaviour gap.
## Verification
- A plugin, module and secret-engine file dropped while the daemon runs loads on
the next tick; an unchanged file is not reloaded.
- An edited file replaces the running interpretation; a deleted file leaves the
running code running, with the process's error naming it until restart.
- A `skills_dir` change refuses the document: the store keeps the version, the
record carries that version and the refusal, and the running `skills_dir` is
unchanged.
- A plugin replacement whose new `Start` fails restarts the old module and
reports the failure; a replacement of a depended-on module is refused and the
dependent keeps running.
- A provider disabled at boot for a missing engine resolves again after the
engine file loads and a tick re-layers.
- The gateway reports the last version it layered at boot, loads no directory,
and its lag appears in its applied-version row rather than the restart banner.
+1 -1
View File
@@ -12,7 +12,7 @@ The dashboard's `POST /api/tasks/{id}/action` (`internal/webui/api_tasks.go`, `h
The path is `handleTaskAction` → `applyOperatorTaskAction` → `Chat.Contract.ApplyOperatorTaskAction` → the daemon's `LocalChatAdapter` → `taskactions.Client` (NATS `archie.gateway.task-action`) → `taskactions.Service.Apply` → one guarded store write that records the decision and requeues. Chat's `/approve` path (`StoreTaskController.Approve` → `chatTaskControllerAdapter.ApproveChatTask`) must reach that same write through the same function, so the two surfaces cannot record different decisions for one operator intent.
A requeue resumes the workflow the waiting state names, never a name the handler hardcodes. `taskactions.Service.apply` and `chatTaskControllerAdapter.ApproveChatTask` requeue every approval under `implement`, which is the feasibility handoff: an approval whose wait recorded `pr-review` would run the implement workflow. Feasibility must keep that handoff by naming `implement` on the task before it enters the wait, which is the invariant the handler needs — the workflow a task carries is the workflow it runs next. `Requeue` treats an empty workflow as "keep the task's".
A requeue resumes the workflow the waiting state names, never a name the handler hardcodes. `taskactions.Service.apply` and `chatTaskControllerAdapter.ApproveChatTask` requeue with an empty workflow, which `Requeue` reads as "keep the task's", so neither surface hardcodes a workflow name: an approval whose wait recorded `pr-review` resumes `pr-review`. The feasibility handoff therefore has to name `implement` on the task before it enters the wait, which is the invariant the handler relies on — the workflow a task carries is the workflow it runs next. `Requeue` treats an empty workflow as "keep the task's".
### The review the operator answers
+4 -1
View File
@@ -103,7 +103,10 @@ Users create and edit workflows in the Web UI. Workflows are stored as versioned
YAML. Built-in workflows can be overridden and restored.
A workflow is a list of steps. Each step has a known type and its own settings.
Plugins can add step types. Each run keeps the workflow version it started with.
A step type is part of the build's step vocabulary, contributed as a provider at
the composition root; interpreted plugins contribute none
(`docs/prds/workflow-step-vocabulary.md`). Each run keeps the workflow version it
started with.
Yaegi, `.archie/stages/*.go`, and `.archie/gate.go` will be removed. Their useful
behaviour must first be available as normal workflow steps.
+3
View File
@@ -58,6 +58,9 @@ same values in Go, and the run's finish check validates them identically.
**Forced by:** a run's only structured results are capture calls
(`agentexec.CaptureTool`, `Result.Captures`), and the workflow interface
already declares when a stage needs them (`task.WorkflowNeeds.Captures`).
`WorkflowInterface.Needs` reports `captures` for a non-empty `Outputs`, so
declaring an output alone requires the capability and a Kit profile that
serves no capture tools is refused before dispatch.
Deriving the tools from the declaration means an output cannot be written
under a name nobody declared, and it keeps one route to structured agent
output rather than adding a second beside the capture path. The row write is
+2 -2
View File
@@ -25,7 +25,7 @@ generic `plugin.Plugin` contract stays metadata-only.
The vocabulary is a compiled-in, closed set. `internal/infrastructure/workflowsteps`
is the one place a step type is bundled into any binary that resolves one: the
shipped workflow stages, the repository-hook replacements (`workflow.diff-rules`)
shipped workflow stages, the repository-hook replacements (`gate.diff-rules`)
and the repository-free event steps each arrive as one provider, and adding a
step type means adding a provider to that set. Interpreted plugin sources
cannot contribute step types: the Yaegi plugin host satisfies only the
@@ -53,7 +53,7 @@ substituting its own registry.
`archie-core-ze7y` is the same question seen from the deleted
repository-authored stages: those `.archie/stages/*.go` files held arbitrary
stages, and their gate half landed as `workflow.diff-rules` settings rather
stages, and their gate half landed as `gate.diff-rules` settings rather
than interpreted Go. Their stage half has no replacement and does not get one
that approximates it: no interpreted-Go stage type is built, and no command
step type is created by this decision to stand in for them. The refusal of
+28
View File
@@ -0,0 +1,28 @@
# The containment workflow a firewall-investigate run calls:
# firewall-investigate (workflow.call, wait: true) hands its findings upward
# from its declared summary, and this workflow's own agent stage decides and
# records the containment as its own declared outputs. Paste it into the
# dashboard's workflow definitions alongside firewall-investigate
# (docs/prds/workflow-call-outputs.md).
id: firewall-contain
repository: none
profile: network-investigator
inputs:
src_ip: { type: string, required: true }
outputs:
contained: { type: bool, required: true }
summary: { type: string }
steps:
- type: workflow.call
settings:
workflow: firewall-investigate
wait: true
inputs: { src_ip: "inputs.src_ip" }
outputs: { summary: "outputs.summary" }
- type: agent.run
settings:
mission: >-
Using the investigation summary the called workflow returned, decide
whether the source address in the inputs should be contained. Write
your decision with the contained tool, and a one-line rationale with
the summary tool.
+8 -1
View File
@@ -2,17 +2,24 @@
# src_ip from a mapped parameter and runs one agent mission in a scratch
# workspace under the network-investigator profile. Paste it into the
# dashboard's workflow definitions.
#
# Its declared output (summary) is offered to the agent stages as a capture
# tool named after it; a wait:true caller publishes it as one of its own
# outputs (docs/prds/workflow-call-outputs.md).
id: firewall-investigate
repository: none
profile: network-investigator
inputs:
src_ip: { type: string, required: true }
severity: { type: string }
outputs:
summary: { type: string }
steps:
- type: agent.run
settings:
mission: >-
Investigate the source address in the inputs: who owns it, what it
resolves to, and whether it is reachable. Report what you found and
how serious it looks given the severity.
how serious it looks given the severity, and write your one-line
conclusion with the summary tool.
read_only: true
+22
View File
@@ -77,6 +77,27 @@ func runGit(t *testing.T, dir string, args ...string) {
}
}
// hostileSigningConfig writes a global git config that makes every commit
// fail unless the fixture disables signing in the repository's own config.
func hostileSigningConfig(t *testing.T) string {
t.Helper()
path := filepath.Join(t.TempDir(), "hostile.gitconfig")
const cfg = "[user]\n\tname = Hostile\n\temail = hostile@example.com\n\tsigningkey = /nonexistent/signing-key\n[gpg]\n\tformat = ssh\n[commit]\n\tgpgsign = true\n"
if err := os.WriteFile(path, []byte(cfg), 0o600); err != nil {
t.Fatal(err)
}
return path
}
// TestFixtureSeedsARepositoryThatIgnoresHostSigningConfig pins the fixture's
// hermeticity: it must not depend on the developer's global git identity or a
// signing agent.
func TestFixtureSeedsARepositoryThatIgnoresHostSigningConfig(t *testing.T) {
t.Setenv("GIT_CONFIG_GLOBAL", hostileSigningConfig(t))
t.Setenv("GIT_CONFIG_NOSYSTEM", "1")
newFixture(t, "edit")
}
type fixture struct {
workspace string
log string
@@ -87,6 +108,7 @@ func newFixture(t *testing.T, scenario string) *fixture {
t.Helper()
ws := t.TempDir()
runGit(t, ws, "init", "-q")
runGit(t, ws, "config", "commit.gpgsign", "false")
if err := os.WriteFile(filepath.Join(ws, "widget_test.go"), []byte("package widget\n"), 0o600); err != nil {
t.Fatal(err)
}
+66
View File
@@ -0,0 +1,66 @@
package agentexec
import (
"errors"
"fmt"
"github.com/samcharles93/ai-sdk/chat"
protocol "github.com/samcharles93/archie-core/internal/tools/mcp"
)
// DefaultSamplingMaxTokens bounds a server-requested completion when the
// request does not set maxTokens. A sampling request is a delegated
// sub-completion, not a full turn, so one MCP server cannot spend a
// turn-sized output allowance on a single request.
const DefaultSamplingMaxTokens = 4096
// SamplingMessages maps a server's sampling messages onto chat messages. Both
// MCP sampling host paths use it -- the daemon's chat handler and the agent
// worker's task handler -- so the same server is answered the same way
// wherever the client runs. A non-text content block or an unknown role is
// rejected rather than silently dropped: answering from a partial prompt would
// be a wrong answer, not a degraded one.
func SamplingMessages(req protocol.SamplingRequest) ([]chat.Message, error) {
if len(req.Messages) == 0 {
return nil, errors.New("mcp sampling: the request has no messages")
}
messages := make([]chat.Message, 0, len(req.Messages))
for i, message := range req.Messages {
if message.Content.Type != "" && message.Content.Type != "text" {
return nil, fmt.Errorf("mcp sampling: message %d has unsupported content type %q", i, message.Content.Type)
}
var role chat.Role
switch message.Role {
case "user":
role = chat.RoleUser
case "assistant":
role = chat.RoleAssistant
default:
return nil, fmt.Errorf("mcp sampling: message %d has unsupported role %q", i, message.Role)
}
messages = append(messages, chat.Message{Role: role, Content: message.Content.Text})
}
return messages, nil
}
// SamplingMaxTokens bounds a delegated completion to the request's own
// maxTokens, falling back to DefaultSamplingMaxTokens, and never past the
// model's own output ceiling (zero when the ceiling is unknown). A
// reasoning-class model gets no bound at all: its provider rejects the
// `max_tokens` parameter the chat-completions provider emits for a non-zero
// bound. Both the daemon's chat path and the agent worker answer sampling
// with this, so the same server gets the same bound on either path.
func SamplingMaxTokens(requested, modelCeiling int, reasoning bool) int {
maxTokens := requested
if maxTokens <= 0 {
maxTokens = DefaultSamplingMaxTokens
}
if modelCeiling > 0 && maxTokens > modelCeiling {
maxTokens = modelCeiling
}
if reasoning {
return 0
}
return maxTokens
}
+98
View File
@@ -0,0 +1,98 @@
package agentexec
import (
"strings"
"testing"
"github.com/samcharles93/ai-sdk/chat"
protocol "github.com/samcharles93/archie-core/internal/tools/mcp"
)
func TestSamplingMessagesMapsRolesAndRejectsUnsupportedInput(t *testing.T) {
tests := []struct {
name string
request protocol.SamplingRequest
want []chat.Message
wantErr string
}{
{
name: "user and assistant text map to chat roles",
request: protocol.SamplingRequest{Messages: []protocol.SamplingMessage{
{Role: "user", Content: protocol.SamplingContent{Type: "text", Text: "2+2?"}},
{Role: "assistant", Content: protocol.SamplingContent{Type: "", Text: "4"}},
}},
want: []chat.Message{
{Role: chat.RoleUser, Content: "2+2?"},
{Role: chat.RoleAssistant, Content: "4"},
},
},
{
name: "no messages is a refusal",
request: protocol.SamplingRequest{},
wantErr: "the request has no messages",
},
{
name: "non-text content is a refusal",
request: protocol.SamplingRequest{Messages: []protocol.SamplingMessage{
{Role: "user", Content: protocol.SamplingContent{Type: "image"}},
}},
wantErr: "unsupported content type",
},
{
name: "unknown role is a refusal",
request: protocol.SamplingRequest{Messages: []protocol.SamplingMessage{
{Role: "system", Content: protocol.SamplingContent{Type: "text", Text: "hi"}},
}},
wantErr: "unsupported role",
},
}
for _, tc := range tests {
t.Run(tc.name, func(t *testing.T) {
got, err := SamplingMessages(tc.request)
if tc.wantErr != "" {
if err == nil || !strings.Contains(err.Error(), tc.wantErr) {
t.Fatalf("SamplingMessages error = %v, want it to mention %q", err, tc.wantErr)
}
return
}
if err != nil {
t.Fatalf("SamplingMessages error = %v, want nil", err)
}
if len(got) != len(tc.want) {
t.Fatalf("SamplingMessages returned %d messages, want %d", len(got), len(tc.want))
}
for i := range got {
if got[i].Role != tc.want[i].Role || got[i].Content != tc.want[i].Content {
t.Errorf("message %d = %+v, want %+v", i, got[i], tc.want[i])
}
}
})
}
}
func TestSamplingMaxTokens(t *testing.T) {
tests := []struct {
name string
requested int
ceiling int
reasoning bool
wantTokens int
}{
{name: "requested bound is honoured", requested: 64, ceiling: 1024, wantTokens: 64},
{name: "absent request falls back to the default", requested: 0, ceiling: 8192, wantTokens: DefaultSamplingMaxTokens},
{name: "request is capped by the model ceiling", requested: 8192, ceiling: 1024, wantTokens: 1024},
{name: "unknown ceiling keeps the request", requested: 8192, ceiling: 0, wantTokens: 8192},
{name: "reasoning model gets no bound", requested: 64, ceiling: 1024, reasoning: true, wantTokens: 0},
{name: "reasoning model with absent request gets no bound", requested: 0, ceiling: 0, reasoning: true, wantTokens: 0},
}
for _, tc := range tests {
t.Run(tc.name, func(t *testing.T) {
if got := SamplingMaxTokens(tc.requested, tc.ceiling, tc.reasoning); got != tc.wantTokens {
t.Fatalf("SamplingMaxTokens(%d, %d, %v) = %d, want %d", tc.requested, tc.ceiling, tc.reasoning, got, tc.wantTokens)
}
})
}
}
+12
View File
@@ -93,6 +93,18 @@ func TestSystemLogHandlerPublishesToTheTasksSystemSubject(t *testing.T) {
if entry.Fields["component"] != "gate" {
t.Errorf("Fields[component] = %v, want gate", entry.Fields["component"])
}
// The wire entry must carry the record's own event time. The daemon files
// a task's records under the time the event happened (see
// logging.TaskRegistry.Write), so a producer that dropped or re-stamped it
// here would collapse a run's timing before it ever left the container.
next.mu.Lock()
wantTime := next.records[0].Time
next.mu.Unlock()
if entry.Time.IsZero() {
t.Error("Entry.Time is the zero time, want the record's own event time")
} else if !entry.Time.Equal(wantTime) {
t.Errorf("Entry.Time = %s, want the record's event time %s", entry.Time, wantTime)
}
}
func TestSystemLogHandlerAlwaysTeesToNextRegardlessOfPublishOutcome(t *testing.T) {
@@ -0,0 +1,93 @@
package agentworker
import (
"log/slog"
"testing"
"github.com/samcharles93/archie-core/internal/agentexec"
"github.com/samcharles93/archie-core/internal/config"
"github.com/samcharles93/archie-core/internal/domain/workflow"
"github.com/samcharles93/archie-core/internal/infrastructure/postgres/pgstore"
"github.com/samcharles93/archie-core/internal/taskrun"
"github.com/samcharles93/archie-core/internal/tools"
)
// guardrailRunYAML is a repository-none workflow with one agent stage: a task
// run reaches AgentStage.handleResult once, and the engine that recorded that
// outcome is observable by its call count and by the decision a low threshold
// earns.
const guardrailRunYAML = "id: guardrail\nrepository: none\nsteps:\n" +
" - type: agent.run\n settings:\n mission: first\n"
// TestProductionWorkerDependenciesBuildGuardrails pins the producer: the
// process that runs agent stages is archie-agent, so its composition root must
// build the guardrail engine every TaskContext records against. A daemon-built
// engine can never reach this process -- it is a separate binary -- so before
// this wiring TaskContext.Guardrails was nil on every production run and no
// decision was ever recorded.
func TestProductionWorkerDependenciesBuildGuardrails(t *testing.T) {
deps, err := productionWorkerDependencies()
if err != nil {
t.Fatalf("productionWorkerDependencies: %v", err)
}
if deps.guardrails == nil {
t.Fatal("worker composition built no guardrail engine: TaskContext.Guardrails stays nil and no agent-stage decision is recorded")
}
}
// TestRunTaskRecordsGuardrailDecision is the end-to-end half: a workflow run
// whose TaskContext carries the engine records its agent stage's outcome. Before
// the wiring the engine was never assigned, so the run recorded zero -- the
// guardrail engine observed nothing for any task.
func TestRunTaskRecordsGuardrailDecision(t *testing.T) {
ctx := t.Context()
st := pgstore.Open(t)
task, err := st.EnqueueChatTask(ctx, "", "", "guardrail recording", "", "guardrail", "", nil)
if err != nil {
t.Fatalf("seed task: %v", err)
}
if err := st.Transition(ctx, task.ID, task.Status, workflow.StatusRunning, "started"); err != nil {
t.Fatalf("claim task: %v", err)
}
task.Status = workflow.StatusRunning
task.WorkflowDefinitionYAML = guardrailRunYAML
task.WorkflowDefinitionDigest = workflow.DigestDefinition(guardrailRunYAML)
if err := st.Update(ctx, task); err != nil {
t.Fatalf("seed task: %v", err)
}
workDir := t.TempDir()
// A threshold a single stage reaches turns the recorded outcome into an
// actual guardrail decision, so the test proves the run consulted the
// engine rather than merely holding a reference.
engine := tools.NewGuardrailEngine(tools.ToolCallGuardrailConfig{
NoProgressWarnAfter: 1,
HardStopAfterWarnRepeat: 1,
})
request := taskrun.Request{
Task: task,
Repo: config.Repo{},
Cfg: config.Config{Models: map[string]string{"builder": "p/m"}}.ForTask(),
WorkflowDefinition: guardrailRunYAML,
}
dependencies := taskDependencies{
store: st,
trees: &fakeScratchTrees{dir: workDir},
steps: testSteps(t),
guardrails: engine,
}
fakeRunner := runnerFactory(func(map[string]agentexec.Provider, *slog.Logger) agentexec.Runner {
return stubRunner{}
})
if _, err := runTask(ctx, request, dependencies, fakeRunner, workDir, slog.New(slog.DiscardHandler)); err != nil {
t.Fatalf("runTask: %v", err)
}
if got := engine.TotalCalls(); got != 1 {
t.Fatalf("guardrail engine recorded %d tool outcomes, want 1 (the agent stage): TaskContext.Guardrails is not wired to the engine the run was given", got)
}
if !engine.HardStopped() {
t.Fatal("guardrail engine issued no decision for the run's agent stage: TaskContext.Guardrails is not wired to the engine the run was given")
}
}
+16 -8
View File
@@ -23,18 +23,26 @@ type mcpProviderSet struct {
registry *tools.Registry
}
type mcpProviderBuilder func(config.MCPServer) (toolprovider.Engine, string, error)
type mcpProviderBuilder func(config.MCPServer, mcp.SamplingHandler) (toolprovider.Engine, string, error)
// startMCPProviders creates MCP transports from config, starts them,
// discovers tools, and registers them into a local registry. Returns
// nil when there are no servers configured.
func startMCPProviders(ctx context.Context, servers []config.MCPServer, log *slog.Logger) (*mcpProviderSet, error) {
return startMCPProvidersWith(ctx, servers, log, buildMCPProvider)
// discovers tools, and registers them into a local registry. A non-nil
// sampling handler answers server-initiated sampling/createMessage requests
// (nil leaves the method refused, the client's own default). Returns nil when
// there are no servers configured.
func startMCPProviders(
ctx context.Context,
servers []config.MCPServer,
sampling mcp.SamplingHandler,
log *slog.Logger,
) (*mcpProviderSet, error) {
return startMCPProvidersWith(ctx, servers, sampling, log, buildMCPProvider)
}
func startMCPProvidersWith(
ctx context.Context,
servers []config.MCPServer,
sampling mcp.SamplingHandler,
log *slog.Logger,
build mcpProviderBuilder,
) (*mcpProviderSet, error) {
@@ -51,7 +59,7 @@ func startMCPProvidersWith(
continue
}
provider, transportType, err := build(server)
provider, transportType, err := build(server, sampling)
if errors.Is(err, errUnknownMCPTransport) {
log.Warn("mcp server skipped: unknown transport", "name", name, "transport", transportType)
continue
@@ -78,7 +86,7 @@ func startMCPProvidersWith(
return set, nil
}
func buildMCPProvider(server config.MCPServer) (toolprovider.Engine, string, error) {
func buildMCPProvider(server config.MCPServer, sampling mcp.SamplingHandler) (toolprovider.Engine, string, error) {
name := strings.TrimSpace(server.Name)
transportType := strings.ToLower(strings.TrimSpace(server.Transport))
if transportType == "" {
@@ -89,7 +97,7 @@ func buildMCPProvider(server config.MCPServer) (toolprovider.Engine, string, err
if err != nil {
return nil, transportType, err
}
return mcptoolprovider.New(name, transport, server.ParallelToolCalls), transportType, nil
return mcptoolprovider.New(name, transport, server.ParallelToolCalls, mcptoolprovider.WithSamplingHandler(sampling)), transportType, nil
}
func mcpTransportForServer(
@@ -167,7 +167,7 @@ func TestBuildMCPProviderSelectsTransport(t *testing.T) {
for _, tc := range tests {
t.Run(tc.name, func(t *testing.T) {
provider, transportType, err := buildMCPProvider(tc.server)
provider, transportType, err := buildMCPProvider(tc.server, nil)
if transportType != tc.wantTransport {
t.Fatalf("transport = %q, want %q", transportType, tc.wantTransport)
}
@@ -344,7 +344,7 @@ func TestStartMCPProvidersPreservesOptionalProviderBehavior(t *testing.T) {
"valid-two": {id: "valid-two", entries: []tools.ToolEntry{testTool("shared")}},
}
buildCalls := make(map[string]int)
build := func(server config.MCPServer) (toolprovider.Engine, string, error) {
build := func(server config.MCPServer, _ protocolmcp.SamplingHandler) (toolprovider.Engine, string, error) {
buildCalls[server.Name]++
switch server.Name {
case "unknown":
@@ -365,7 +365,7 @@ func TestStartMCPProvidersPreservesOptionalProviderBehavior(t *testing.T) {
{Name: "valid-one"},
{Name: "valid-two"},
}
set, err := startMCPProvidersWith(t.Context(), servers, slog.New(slog.DiscardHandler), build)
set, err := startMCPProvidersWith(t.Context(), servers, nil, slog.New(slog.DiscardHandler), build)
if err != nil {
t.Fatalf("error = %v, want nil when valid providers remain available", err)
}
@@ -409,14 +409,14 @@ func TestStartMCPProvidersReturnsNilWithoutAvailableProviders(t *testing.T) {
{
name: "unknown transport is skipped without error",
servers: []config.MCPServer{{Name: "unknown"}},
build: func(config.MCPServer) (toolprovider.Engine, string, error) {
build: func(config.MCPServer, protocolmcp.SamplingHandler) (toolprovider.Engine, string, error) {
return nil, "unknown", errUnknownMCPTransport
},
},
{
name: "first provider error is returned",
servers: []config.MCPServer{{Name: "first"}, {Name: "second"}},
build: func(server config.MCPServer) (toolprovider.Engine, string, error) {
build: func(server config.MCPServer, _ protocolmcp.SamplingHandler) (toolprovider.Engine, string, error) {
if server.Name == "first" {
return nil, "stdio", firstErr
}
@@ -430,12 +430,12 @@ func TestStartMCPProvidersReturnsNilWithoutAvailableProviders(t *testing.T) {
t.Run(tc.name, func(t *testing.T) {
build := tc.build
if build == nil {
build = func(config.MCPServer) (toolprovider.Engine, string, error) {
build = func(config.MCPServer, protocolmcp.SamplingHandler) (toolprovider.Engine, string, error) {
t.Fatal("builder called without configured servers")
return nil, "", nil
}
}
set, err := startMCPProvidersWith(t.Context(), tc.servers, slog.New(slog.DiscardHandler), build)
set, err := startMCPProvidersWith(t.Context(), tc.servers, nil, slog.New(slog.DiscardHandler), build)
if set != nil {
t.Fatalf("provider set = %#v, want nil", set)
}
+67
View File
@@ -0,0 +1,67 @@
package agentworker
import (
"context"
"errors"
"fmt"
"strings"
"github.com/samcharles93/ai-sdk/core"
"github.com/samcharles93/ai-sdk/runtime"
"github.com/samcharles93/archie-core/internal/agentexec"
"github.com/samcharles93/archie-core/internal/config"
"github.com/samcharles93/archie-core/internal/tools/mcp"
)
// taskSamplingModel is the model a task-scoped MCP server's sampling request
// is answered with: the task's builder model, the same fallback an agent stage
// whose role is unset uses (workflow.AgentStage.resolveModel) and the same
// role the daemon's chat path falls back to. Empty means the task has no model
// to answer with.
func taskSamplingModel(cfg config.TaskConfig) string {
return strings.TrimSpace(cfg.Models["builder"])
}
// taskSamplingHandler answers an MCP server's sampling/createMessage request
// from the task's own model runtime, the same way the daemon's chat path
// answers it from its chat model. The worker hosts the same configured MCP
// servers as a chat turn, so without this a server's sampling request is
// answered on one path and refused on the other. A missing runtime or model is
// a refused request, never a dropped one: the client maps the returned error
// to a JSON-RPC error for that request alone.
func taskSamplingHandler(llm *runtime.Runtime, cfg config.TaskConfig) mcp.SamplingHandler {
return func(ctx context.Context, req mcp.SamplingRequest) (mcp.SamplingResult, error) {
if llm == nil {
return mcp.SamplingResult{}, errors.New("mcp sampling: no model runtime is configured")
}
model := taskSamplingModel(cfg)
if model == "" {
return mcp.SamplingResult{}, errors.New("mcp sampling: no builder model is configured")
}
messages, err := agentexec.SamplingMessages(req)
if err != nil {
return mcp.SamplingResult{}, err
}
limits := cfg.ModelLimits[model]
options := core.GenerateOptions{
Messages: messages,
System: req.SystemPrompt,
MaxSteps: 1,
MaxTokens: agentexec.SamplingMaxTokens(req.MaxTokens, limits.MaxOutputTokens, limits.Reasoning),
}
if req.Temperature != nil {
options.Temperature = float32(*req.Temperature)
}
result, err := llm.Chat(ctx, model, options)
if err != nil {
return mcp.SamplingResult{}, fmt.Errorf("mcp sampling: %w", err)
}
return mcp.SamplingResult{
Model: model,
Role: "assistant",
Content: mcp.SamplingContent{Type: "text", Text: result.Text},
StopReason: string(result.FinishReason),
}, nil
}
}
@@ -0,0 +1,193 @@
package agentworker
import (
"encoding/json"
"io"
"net/http"
"net/http/httptest"
"reflect"
"strings"
"sync"
"testing"
"github.com/samcharles93/archie-core/internal/agentexec"
"github.com/samcharles93/archie-core/internal/config"
"github.com/samcharles93/archie-core/internal/tools/mcp"
)
// capturedChatRequest is the JSON body the OpenAI-compatible provider sent for
// the last call, which is where the max-token parameter is chosen.
type capturedChatRequest struct {
mu sync.Mutex
body map[string]any
}
func (c *capturedChatRequest) get() map[string]any {
c.mu.Lock()
defer c.mu.Unlock()
return c.body
}
// newCapturingChatAPI answers every completion and records the wire body, so a
// test can assert the parameter the provider actually emitted rather than an
// option the handler merely set.
func newCapturingChatAPI(t *testing.T, capture *capturedChatRequest) *httptest.Server {
t.Helper()
api := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
var body map[string]any
_ = json.NewDecoder(r.Body).Decode(&body)
capture.mu.Lock()
capture.body = body
capture.mu.Unlock()
w.Header().Set("Content-Type", "application/json")
_, _ = io.WriteString(w, `{"id":"chatcmpl-1","object":"chat.completion","created":1,"model":"test",`+
`"choices":[{"index":0,"message":{"role":"assistant","content":"ok"},"finish_reason":"stop"}],`+
`"usage":{"prompt_tokens":1,"completion_tokens":1,"total_tokens":2}}`)
}))
t.Cleanup(api.Close)
return api
}
// newSamplingModelServer stands in for an OpenAI-compatible provider so a
// sampling request can be answered without a real backend.
func newSamplingModelServer(t *testing.T) *httptest.Server {
t.Helper()
api := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {
w.Header().Set("Content-Type", "application/json")
_, _ = io.WriteString(w, `{"id":"chatcmpl-1","object":"chat.completion","created":1,"model":"gpt-5.6",`+
`"choices":[{"index":0,"message":{"role":"assistant","content":"four"},"finish_reason":"stop"}],`+
`"usage":{"prompt_tokens":1,"completion_tokens":1,"total_tokens":2}}`)
}))
t.Cleanup(api.Close)
return api
}
func testSamplingRequest() mcp.SamplingRequest {
return mcp.SamplingRequest{
Messages: []mcp.SamplingMessage{{
Role: "user",
Content: mcp.SamplingContent{Type: "text", Text: "2+2?"},
}},
MaxTokens: 16,
}
}
// TestTaskSamplingHandlerAnswersFromWorkerModel proves a task-scoped MCP
// server's sampling request is answered by the worker's own model. A task
// hosts the same configured MCP servers as a chat turn, so without this the
// server is answered on the chat path and refused on the task path.
func TestTaskSamplingHandlerAnswersFromWorkerModel(t *testing.T) {
api := newSamplingModelServer(t)
llm := agentexec.NewRuntime(map[string]agentexec.Provider{
"openai": {Class: "openai", BaseURL: api.URL},
})
handler := taskSamplingHandler(llm, config.TaskConfig{Models: map[string]string{"builder": "openai/gpt-5.6"}})
result, err := handler(t.Context(), testSamplingRequest())
if err != nil {
t.Fatalf("sampling handler = %v, want a completion", err)
}
if result.Model != "openai/gpt-5.6" {
t.Errorf("result.Model = %q, want the task's builder model", result.Model)
}
if result.Role != "assistant" || result.Content.Type != "text" || result.Content.Text != "four" {
t.Errorf("result = %+v, want an assistant text completion", result)
}
}
// TestTaskSamplingHandlerRefusesCleanly proves a worker with no usable model
// answers the request with an error instead of panicking or dropping it: the
// server sees a failed request, not a dead session.
func TestTaskSamplingHandlerRefusesCleanly(t *testing.T) {
tests := []struct {
name string
llm bool
model string
want string
}{
{name: "no model runtime", llm: false, model: "openai/gpt-5.6", want: "no model runtime"},
{name: "no builder model configured", llm: true, want: "no builder model"},
}
for _, tc := range tests {
t.Run(tc.name, func(t *testing.T) {
llm := agentexec.NewRuntime(nil)
if tc.llm {
llm = agentexec.NewRuntime(map[string]agentexec.Provider{"openai": {Class: "openai"}})
}
models := map[string]string{}
if tc.model != "" {
models["builder"] = tc.model
}
_, err := taskSamplingHandler(llm, config.TaskConfig{Models: models})(t.Context(), testSamplingRequest())
if err == nil {
t.Fatal("sampling handler returned no error with no usable model")
}
if !strings.Contains(err.Error(), tc.want) {
t.Errorf("error = %q, want it to mention %q", err, tc.want)
}
})
}
}
// TestTaskSamplingBoundFollowsModelClass proves the worker consumes the
// reasoning flag carried in TaskConfig.ModelLimits: a reasoning-class builder
// model gets no max-token bound (its provider rejects the max_tokens parameter
// the chat-completions provider emits), while a classic model gets the
// request's own bound.
func TestTaskSamplingBoundFollowsModelClass(t *testing.T) {
tests := []struct {
name string
reasoning bool
wantBound bool
wantValue int
}{
{name: "reasoning model", reasoning: true},
{name: "classic model", wantBound: true, wantValue: 16},
}
for _, tc := range tests {
t.Run(tc.name, func(t *testing.T) {
capture := &capturedChatRequest{}
api := newCapturingChatAPI(t, capture)
llm := agentexec.NewRuntime(map[string]agentexec.Provider{
"openai": {Class: "openai", BaseURL: api.URL},
})
cfg := config.TaskConfig{
Models: map[string]string{"builder": "openai/gpt-5.6"},
ModelLimits: map[string]config.ModelLimits{
"openai/gpt-5.6": {MaxOutputTokens: 128000, Reasoning: tc.reasoning},
},
}
if _, err := taskSamplingHandler(llm, cfg)(t.Context(), testSamplingRequest()); err != nil {
t.Fatalf("sampling handler error = %v", err)
}
bound, present := capture.get()["max_tokens"]
if present != tc.wantBound {
t.Fatalf("request body max_tokens present = %v (value %v), want %v; body = %v", present, bound, tc.wantBound, capture.get())
}
if present && bound != float64(tc.wantValue) {
t.Fatalf("request body max_tokens = %v, want %d; body = %v", bound, tc.wantValue, capture.get())
}
})
}
}
// TestBuildMCPProviderWiresTaskSamplingHandler proves the worker's MCP
// provider builder hands the provider the sampling handler, so the seam a task
// run builds providers through is the one that answers sampling -- not only
// the handler in isolation.
func TestBuildMCPProviderWiresTaskSamplingHandler(t *testing.T) {
handler := taskSamplingHandler(agentexec.NewRuntime(map[string]agentexec.Provider{
"openai": {Class: "openai"},
}), config.TaskConfig{Models: map[string]string{"builder": "openai/gpt-5.6"}})
provider, _, err := buildMCPProvider(config.MCPServer{Name: "sampling", Command: "server"}, handler)
if err != nil {
t.Fatalf("buildMCPProvider: %v", err)
}
field := reflect.ValueOf(provider).Elem().FieldByName("samplingHandler")
if field.IsNil() {
t.Fatal("task MCP provider carries no sampling handler: a task-scoped server's sampling/createMessage request is refused")
}
}
+42 -33
View File
@@ -10,6 +10,8 @@ import (
"path/filepath"
"strconv"
"github.com/samcharles93/ai-sdk/runtime"
"github.com/samcharles93/archie-core/internal/agentexec"
"github.com/samcharles93/archie-core/internal/config"
"github.com/samcharles93/archie-core/internal/domain/workflow"
@@ -19,6 +21,7 @@ import (
"github.com/samcharles93/archie-core/internal/installtype"
"github.com/samcharles93/archie-core/internal/storage"
"github.com/samcharles93/archie-core/internal/taskrun"
"github.com/samcharles93/archie-core/internal/tools"
"github.com/samcharles93/archie-core/internal/worktree"
)
@@ -48,7 +51,15 @@ type hybridTrees struct {
worktreeUID, worktreeGID int
}
func (h *hybridTrees) Prepare(ctx context.Context, owner, repo, base string, issue int, title, body, labels string) (dir, branch string, err error) {
// Prepare binds the task's already-prepared, bind-mounted worktree. The
// daemon positioned it before the container started, so this resolves the
// directory and branch rather than doing network work. A resume target (the
// PR branch a remediation continues) is returned verbatim; a fresh target uses
// the branch the daemon computed and persisted on the task row.
func (h *hybridTrees) Prepare(ctx context.Context, owner, repo, base string, issue int, title, body, labels string, target workflow.PrepareTarget) (dir, branch string, err error) {
if target != workflow.PrepareFresh {
return h.localDir, string(target), nil
}
return h.localDir, h.branch, nil
}
@@ -107,22 +118,6 @@ func (h *hybridTrees) Snapshot(ctx context.Context, dir, destDir string) error {
return h.local.Snapshot(ctx, dir, destDir)
}
// Resume is a no-op here for the same reason Prepare is: resuming a PR
// branch needs the daemon's forge credential to fetch, so archied must run
// it against the bind-mounted worktree before the container starts, the
// same way it runs Prepare today. The container's worktree is already
// resumed by the time a remediate run reaches this stage.
func (h *hybridTrees) Resume(ctx context.Context, dir, branch string) error {
return nil
}
// Dir returns the bind-mounted worktree path archied already prepared,
// ignoring its arguments -- the sandbox has no way to independently derive
// or verify the path, same as Prepare above.
func (h *hybridTrees) Dir(owner, repo string, issue int) string {
return h.localDir
}
func (h *hybridTrees) ChangedLines(ctx context.Context, dir, base string) (int, error) {
return h.local.ChangedLines(ctx, dir, base)
}
@@ -192,6 +187,11 @@ type taskDependencies struct {
// executing half of the contract whose validating half is the control
// plane's (archie-core-fwmp).
steps *workflow.Manager
// guardrails is the guardrail engine agent stages record tool outcomes
// against. It is built by this process's composition root (the workflow
// runs in archie-agent, so the daemon's engine cannot reach it) and set on
// every TaskContext; nil means guardrails are disabled.
guardrails *tools.GuardrailEngine
}
type runnerFactory func(map[string]agentexec.Provider, *slog.Logger) agentexec.Runner
@@ -218,15 +218,17 @@ func applyToolLimits(agent agentexec.Runner, policy config.ToolPolicy, allow []s
runner.AllowTools = allow
}
// stageRunners builds the runner every agent stage uses.
func stageRunners(req taskrun.Request, mcpSet *mcpProviderSet, newRunner runnerFactory, log *slog.Logger) (agentexec.Runner, error) {
// stageRunners builds the runner every agent stage uses. llm is the task's
// shared model runtime (the one its MCP providers answer sampling from); the
// registry path reuses it rather than building a second one.
func stageRunners(req taskrun.Request, mcpSet *mcpProviderSet, newRunner runnerFactory, llm *runtime.Runtime, log *slog.Logger) (agentexec.Runner, error) {
if req.Harness != nil {
// The built-in loop has no route to a model from a Kit container.
return agentexec.HarnessStages{Runner: agentexec.NewHarnessRunner(nil), Spec: *req.Harness}, nil
}
var agent agentexec.Runner
if mcpSet != nil && mcpSet.registry != nil {
agent = agentexec.NewLoopRunner(agentexec.NewRuntime(req.Providers), log, mcpSet.registry)
agent = agentexec.NewLoopRunner(llm, log, mcpSet.registry)
} else {
agent = newRunner(req.Providers, log)
}
@@ -277,8 +279,14 @@ func runTask(ctx context.Context, req taskrun.Request, dependencies taskDependen
// start from.
defer restoreWorktreeOwnership(trees, workDir, log)()
// Start MCP providers and build a local tool registry.
mcpSet, mcpErr := startMCPProviders(ctx, req.MCPServers, log)
// The MCP providers a task hosts share the same model runtime its agent
// stages run on, so a server-initiated sampling request is answered from
// the task's own model rather than a second provider client.
var llm *runtime.Runtime
if len(req.MCPServers) > 0 {
llm = agentexec.NewRuntime(req.Providers)
}
mcpSet, mcpErr := startMCPProviders(ctx, req.MCPServers, taskSamplingHandler(llm, req.Cfg), log)
if mcpSet != nil {
defer mcpSet.cleanup(ctx, log)
}
@@ -286,7 +294,7 @@ func runTask(ctx context.Context, req taskrun.Request, dependencies taskDependen
log.Warn("mcp providers had errors, continuing with available tools", "err", mcpErr)
}
agent, err := stageRunners(req, mcpSet, newRunner, log)
agent, err := stageRunners(req, mcpSet, newRunner, llm, log)
if err != nil {
return nil, err
}
@@ -310,16 +318,17 @@ func runTask(ctx context.Context, req taskrun.Request, dependencies taskDependen
}
tc := &workflow.TaskContext{
Task: req.Task,
Repo: req.Repo,
Cfg: req.Cfg.ToConfig(),
Forge: dependencies.forge,
Store: dependencies.store,
Calls: dependencies.calls,
Trees: trees,
Agent: agent,
Bus: bus,
Log: log,
Task: req.Task,
Repo: req.Repo,
Cfg: req.Cfg.ToConfig(),
Forge: dependencies.forge,
Store: dependencies.store,
Calls: dependencies.calls,
Trees: trees,
Agent: agent,
Bus: bus,
Log: log,
Guardrails: dependencies.guardrails,
}
// pr-review reads an external pull request's data from workDir, which the
// daemon already populated before this task started (daemon's
@@ -170,6 +170,31 @@ func TestHybridTreesPushSkipsReconciliationWhenNotConfigured(t *testing.T) {
}
}
// TestHybridTreesPrepareResolvesTheTargetBranch pins the container half of the
// workflow prepare seam: the daemon has already positioned the bind-mounted
// worktree, so the adapter resolves its directory and returns the requested
// resume target verbatim -- the branch a remediation continues -- while a fresh
// prepare returns the branch the daemon persisted on the task row.
func TestHybridTreesPrepareResolvesTheTargetBranch(t *testing.T) {
trees := &hybridTrees{localDir: "/work/issue-1", branch: "archie/fresh"}
dir, branch, err := trees.Prepare(t.Context(), "o", "r", "main", 1, "title", "", "", workflow.PrepareTarget("archie/pr-branch"))
if err != nil {
t.Fatalf("Prepare(resume): %v", err)
}
if dir != "/work/issue-1" || branch != "archie/pr-branch" {
t.Fatalf("Prepare(resume) = %q/%q, want /work/issue-1/archie/pr-branch", dir, branch)
}
dir, branch, err = trees.Prepare(t.Context(), "o", "r", "main", 1, "title", "", "", workflow.PrepareFresh)
if err != nil {
t.Fatalf("Prepare(fresh): %v", err)
}
if dir != "/work/issue-1" || branch != "archie/fresh" {
t.Fatalf("Prepare(fresh) = %q/%q, want /work/issue-1/archie/fresh", dir, branch)
}
}
func TestHybridTreesCommitReconcilesOwnershipEvenWhenCommitFails(t *testing.T) {
missing := filepath.Join(t.TempDir(), "does-not-exist")
trees := &hybridTrees{
@@ -301,8 +326,8 @@ type remoteManager struct {
branch string
}
func (r *remoteManager) Prepare(ctx context.Context, owner, repo, base string, issue int, title, body, labels string) (string, string, error) {
dir, branch, err := r.manager.Prepare(ctx, owner, repo, base, issue, title, body, labels)
func (r *remoteManager) Prepare(ctx context.Context, owner, repo, base string, issue int, title, body, labels string, target workflow.PrepareTarget) (string, string, error) {
dir, branch, err := r.manager.Prepare(ctx, owner, repo, base, issue, title, body, labels, worktree.Target(target))
if err == nil {
r.dir = dir
r.branch = branch
@@ -338,6 +363,7 @@ func newLocalRemote(t *testing.T, owner, repo string) string {
run(seed, "init", "-b", "main")
run(seed, "config", "user.name", "seeder")
run(seed, "config", "user.email", "seed@example.com")
run(seed, "config", "commit.gpgsign", "false")
if err := os.WriteFile(filepath.Join(seed, "README.md"), []byte("seed\n"), 0o644); err != nil {
t.Fatal(err)
}
@@ -347,6 +373,26 @@ func newLocalRemote(t *testing.T, owner, repo string) string {
return host
}
// hostileSigningConfig writes a global git config that makes every commit
// fail unless the fixture disables signing in the repository's own config.
func hostileSigningConfig(t *testing.T) string {
t.Helper()
path := filepath.Join(t.TempDir(), "hostile.gitconfig")
const cfg = "[user]\n\tname = Hostile\n\temail = hostile@example.com\n\tsigningkey = /nonexistent/signing-key\n[gpg]\n\tformat = ssh\n[commit]\n\tgpgsign = true\n"
if err := os.WriteFile(path, []byte(cfg), 0o600); err != nil {
t.Fatal(err)
}
return path
}
// TestNewLocalRemoteIgnoresHostSigningConfig pins the fixture's hermeticity:
// it must not depend on the developer's global git identity or a signing agent.
func TestNewLocalRemoteIgnoresHostSigningConfig(t *testing.T) {
t.Setenv("GIT_CONFIG_GLOBAL", hostileSigningConfig(t))
t.Setenv("GIT_CONFIG_NOSYSTEM", "1")
newLocalRemote(t, "acme", "todo")
}
func startEmbeddedTaskRPCServer(t *testing.T) *server.Server {
t.Helper()
srv := natssrv.RunRandClientPortServer()
@@ -407,7 +453,7 @@ func TestRunTaskExecutesBootstrapWorkflowEndToEnd(t *testing.T) {
fg := &prCapturingForge{}
manager := &worktree.Manager{WorkDir: t.TempDir(), Token: "unused", BotUser: "archie-bot", BotEmail: "archie-bot@example.com", BaseURL: "file://" + host}
remote := &remoteManager{manager: manager}
hostDir, _, err := remote.Prepare(ctx, "acme", "widget", "main", 1, task.Title, "", "bootstrap")
hostDir, _, err := remote.Prepare(ctx, "acme", "widget", "main", 1, task.Title, "", "bootstrap", workflow.PrepareFresh)
if err != nil {
t.Fatalf("prepare: %v", err)
}
@@ -613,7 +659,7 @@ func TestExecuteTaskRequestUsesInfrastructureRPCDependencies(t *testing.T) {
BotEmail: "archie-bot@example.com",
BaseURL: "file://" + host,
}
hostDir, branch, err := daemonTrees.Prepare(ctx, "acme", "rpc-widget", "main", 2, task.Title, "", "bootstrap")
hostDir, branch, err := daemonTrees.Prepare(ctx, "acme", "rpc-widget", "main", 2, task.Title, "", "bootstrap", worktree.Fresh)
if err != nil {
t.Fatal(err)
}
@@ -650,7 +696,7 @@ func TestExecuteTaskRequestUsesInfrastructureRPCDependencies(t *testing.T) {
Cfg: config.Config{}.ForTask(),
WorktreeGrant: grant,
}
response, err := executeTaskRequest(ctx, request, transport, hostDir, slog.New(slog.DiscardHandler), testSteps(t))
response, err := executeTaskRequest(ctx, request, transport, hostDir, slog.New(slog.DiscardHandler), testSteps(t), nil)
if err != nil {
t.Fatal(err)
}
@@ -705,7 +751,7 @@ func TestExecuteTaskRequestForwardsWorkflowEventsOverNATS(t *testing.T) {
BotEmail: "archie-bot@example.com",
BaseURL: "file://" + host,
}
hostDir, branch, err := daemonTrees.Prepare(ctx, "acme", "events-widget", "main", 3, task.Title, "", "bootstrap")
hostDir, branch, err := daemonTrees.Prepare(ctx, "acme", "events-widget", "main", 3, task.Title, "", "bootstrap", worktree.Fresh)
if err != nil {
t.Fatal(err)
}
@@ -768,7 +814,7 @@ func TestExecuteTaskRequestForwardsWorkflowEventsOverNATS(t *testing.T) {
Cfg: config.Config{}.ForTask(),
WorktreeGrant: grant,
}
response, err := executeTaskRequest(ctx, request, transport, hostDir, slog.New(slog.DiscardHandler), testSteps(t))
response, err := executeTaskRequest(ctx, request, transport, hostDir, slog.New(slog.DiscardHandler), testSteps(t), nil)
if err != nil {
t.Fatal(err)
}
@@ -898,7 +944,7 @@ func TestHybridTreesForwardsChangedFileStatsToItsLocalManager(t *testing.T) {
WorkDir: t.TempDir(), Token: "unused",
BotUser: "archie-bot", BotEmail: "archie-bot@example.com", BaseURL: "file://" + host,
}
dir, branch, err := manager.Prepare(ctx, "acme", "diffstat", "main", 4, "feat: widget", "", "")
dir, branch, err := manager.Prepare(ctx, "acme", "diffstat", "main", 4, "feat: widget", "", "", worktree.Fresh)
if err != nil {
t.Fatalf("prepare: %v", err)
}
@@ -34,7 +34,7 @@ type fakeScratchTrees struct {
dir string
}
func (f *fakeScratchTrees) Prepare(context.Context, string, string, string, int, string, string, string) (string, string, error) {
func (f *fakeScratchTrees) Prepare(context.Context, string, string, string, int, string, string, string, workflow.PrepareTarget) (string, string, error) {
return f.dir, "", nil
}
+17 -9
View File
@@ -15,6 +15,7 @@ import (
"github.com/samcharles93/archie-core/internal/infrastructure/workflowsteps"
"github.com/samcharles93/archie-core/internal/storage"
"github.com/samcharles93/archie-core/internal/taskrun"
"github.com/samcharles93/archie-core/internal/tools"
)
const rpcTimeout = 60 * time.Second
@@ -64,6 +65,11 @@ type workerDependencies struct {
// definitions against, registered by the composition root before the first
// task is served (infrastructure/workflowsteps is the provider set).
steps *workflow.Manager
// guardrails is the guardrail engine every served task records agent-stage
// outcomes against, built once at this process's composition root.
// archie-agent is the process that runs the stages, so the engine that
// observes them lives here; a live engine cannot cross the task handoff.
guardrails *tools.GuardrailEngine
}
// productionWorkerDependencies builds the worker's process dependencies. The
@@ -87,7 +93,8 @@ func productionWorkerDependencies() (workerDependencies, error) {
<-ctx.Done()
log.Info("archie-agent shutting down")
},
steps: steps,
steps: steps,
guardrails: tools.NewGuardrailEngine(tools.DefaultGuardrailConfig()),
}, nil
}
@@ -134,7 +141,7 @@ func run(ctx context.Context, settings Settings, log *slog.Logger, dependencies
log.Info("system log publisher attached", "task", taskID)
subscription, err := transport.SubscribeTasks(ctx, taskID, func(ctx context.Context, request taskrun.Request) (*taskrun.Response, error) {
return executeTaskRequest(ctx, request, transport, workDir, log, dependencies.steps)
return executeTaskRequest(ctx, request, transport, workDir, log, dependencies.steps, dependencies.guardrails)
}, log)
if err != nil {
log.Error("taskrun subscribe failed", "err", err)
@@ -159,15 +166,16 @@ type taskServiceTransport interface {
EventPublisher() agentexec.EventPublisher
}
func executeTaskRequest(ctx context.Context, request taskrun.Request, transport taskServiceTransport, workDir string, log *slog.Logger, steps *workflow.Manager) (*taskrun.Response, error) {
func executeTaskRequest(ctx context.Context, request taskrun.Request, transport taskServiceTransport, workDir string, log *slog.Logger, steps *workflow.Manager, guardrails *tools.GuardrailEngine) (*taskrun.Response, error) {
log.Info("running task", "task", request.Task.ID, "repo", request.Repo.FullName(), "issue", request.Task.IssueNumber)
dependencies := taskDependencies{
forge: transport.Forger(request.Task.Identity, rpcTimeout),
store: transport.Store(rpcTimeout),
calls: transport.Calls(rpcTimeout),
trees: transport.Trees(request.Task.Identity, request.WorktreeGrant, rpcTimeout),
events: transport.EventPublisher(),
steps: steps,
forge: transport.Forger(request.Task.Identity, rpcTimeout),
store: transport.Store(rpcTimeout),
calls: transport.Calls(rpcTimeout),
trees: transport.Trees(request.Task.Identity, request.WorktreeGrant, rpcTimeout),
events: transport.EventPublisher(),
steps: steps,
guardrails: guardrails,
}
response, err := runTask(ctx, request, dependencies, newTaskRunner, workDir, log)
if err != nil {
+245 -55
View File
@@ -17,6 +17,7 @@ import (
"path/filepath"
"slices"
"strings"
"sync"
"sync/atomic"
"syscall"
"time"
@@ -31,6 +32,7 @@ import (
controlpb "github.com/samcharles93/archie-core/internal/contracts/controlplane/v1"
"github.com/samcharles93/archie-core/internal/daemon"
"github.com/samcharles93/archie-core/internal/domain/access"
"github.com/samcharles93/archie-core/internal/domain/agent"
"github.com/samcharles93/archie-core/internal/domain/applystatus"
"github.com/samcharles93/archie-core/internal/domain/curator"
"github.com/samcharles93/archie-core/internal/domain/eda/module"
@@ -39,6 +41,7 @@ import (
"github.com/samcharles93/archie-core/internal/domain/health"
"github.com/samcharles93/archie-core/internal/domain/identity"
domainmemory "github.com/samcharles93/archie-core/internal/domain/memory"
"github.com/samcharles93/archie-core/internal/domain/messaging"
"github.com/samcharles93/archie-core/internal/domain/scheduling"
"github.com/samcharles93/archie-core/internal/domain/storecontract"
"github.com/samcharles93/archie-core/internal/domain/workflow"
@@ -59,6 +62,7 @@ import (
"github.com/samcharles93/archie-core/internal/infrastructure/staterpc"
"github.com/samcharles93/archie-core/internal/infrastructure/taskactions"
"github.com/samcharles93/archie-core/internal/infrastructure/toolbuilder"
"github.com/samcharles93/archie-core/internal/infrastructure/transcription"
"github.com/samcharles93/archie-core/internal/logging"
"github.com/samcharles93/archie-core/internal/plugin"
"github.com/samcharles93/archie-core/internal/plugin/pluginextract"
@@ -165,8 +169,22 @@ type boot struct {
// closes; the Gateway also holds its serve claim on it.
chatPool *pgxpool.Pool
// catalog, catalogModels and catalogMu are the model catalog the process
// last loaded and the model references it contributes. The refresh loop
// replaces both together while the config paths read them, so they are one
// value behind one lock rather than two fields that can disagree.
catalogMu sync.RWMutex
catalog modelcatalog.Snapshot
catalogModels []string
// catalogCachePath is where the catalog's last download is cached, beside
// the config file. Empty when boot had no file path to derive it from.
catalogCachePath string
// catalogURL overrides the catalog endpoint. Zero in production:
// modelcatalog defaults to models.dev. Set only by tests.
catalogURL string
// catalogRefreshInterval is how often the running daemon re-reads the
// catalog. The daemon's refresh loop reads it; zero means the default.
catalogRefreshInterval time.Duration
bus *events.Bus
// taskActionsConn is the standalone Gateway process's own NATS connection:
@@ -238,6 +256,15 @@ type boot struct {
// unrelated call.
embeddings domainembedding.Client
// transcriber is the optional voice-transcription capability. It is built
// beside the chat runtime from this process's own [models]/[providers],
// which is the model-owning side of the Messaging boundary: a channel
// frontend carries a voice note's bytes across the inbound wire, and this
// process turns them into text before the turn is recorded. Nil when no
// role is configured or the provider credential did not resolve; a turn
// then keeps the frontend's media note rather than failing.
transcriber messaging.Transcriber
toolReg *tools.Registry
chatModels *chatModelManager
personas *gateway.PersonaRegistry
@@ -246,17 +273,34 @@ type boot struct {
defaultChatIdentity string
updateService *releaseupdate.Service
startGateways []func()
capabilityHost *plugin.Host
startGateways []func()
capabilityHost *plugin.Host
// pluginReconciler loads plugin, module and secret-engine files dropped into
// the running config's directories without a restart. Nil before boot's
// startPluginReconcile, and in processes that load no directory (the
// Gateway), where a stored plugin-settings change still layers but loads
// nothing.
pluginReconciler *pluginReconciler
trees *worktree.Manager
worktreeGrants *worktreerpc.Grants
identityRunners []*daemon.IdentityRunner
memEngines *domainmemory.Registry
curatorRegistry *curator.Registry
curatorRuntime *curator.Runtime
guardrails *tools.GuardrailEngine
providerRegistry *toolprovider.Registry
d *daemon.Daemon
// mcpMu guards mcpApplied, which records the MCP servers the provider
// registry was last built from, keyed by configured name. A live
// tool-settings change diffs the stored server set against it so an
// unchanged server is left running and untouched (mcp_reconcile.go); the
// mutex is held because registerTools runs at boot while the
// control-plane watch may already be applying an update.
mcpMu sync.Mutex
mcpApplied map[string]appliedMCPServer
// mcpProvider builds the engine for one configured MCP server. Nil uses
// configuredMCPProvider; a test injects a fake so the live reconciliation
// can run without spawning a server process.
mcpProvider func(config.MCPServer) (toolprovider.Engine, error)
d *daemon.Daemon
// schedulingEngine is the cron/scheduling ticker engine (setupScheduling).
// Nil when no chat task creator is configured, in which case startServices
// leaves it unstarted rather than running with no reachable job kind.
@@ -530,19 +574,146 @@ func (b *boot) handleRequeue(ctx context.Context, requeue int64, once bool) (boo
return !once, nil
}
// catalogState returns the model catalog the process is running and the model
// references it contributes. Both together, under one lock: the refresh loop
// replaces them as a pair.
func (b *boot) catalogState() (modelcatalog.Snapshot, []string) {
b.catalogMu.RLock()
defer b.catalogMu.RUnlock()
return b.catalog, b.catalogModels
}
// setCatalogState installs a loaded catalog. Called only after everything it
// has to reach has been built, so a read that fails leaves the previous
// snapshot running.
func (b *boot) setCatalogState(snapshot modelcatalog.Snapshot, models []string) {
b.catalogMu.Lock()
defer b.catalogMu.Unlock()
b.catalog = snapshot
b.catalogModels = models
}
// loadCatalog reads the model catalog once at boot and merges it into the
// running config. A catalog that cannot be read is not fatal -- the configured
// providers and models keep running, and the refresh loop retries the same
// read -- but a catalog that *is* read must reach the config the process runs
// on, not just the boot struct the chat runtime was built from.
func (b *boot) loadCatalog(ctx context.Context, cfgPath string) {
catalog, err := modelcatalog.Load(ctx, modelcatalog.Options{
CachePath: filepath.Join(filepath.Dir(cfgPath), "models.json"),
Getenv: b.secrets.Getenv,
Configured: b.cfg.Providers,
})
b.catalog = catalog
b.catalogCachePath = filepath.Join(filepath.Dir(cfgPath), "models.json")
catalog, err := modelcatalog.Load(ctx, b.catalogOptions(b.catalogCachePath))
if err != nil {
b.log.Warn("model catalog unavailable; using configured providers and models", "err", err)
return
}
b.catalogModels = applyModelCatalog(&b.cfg, catalog)
b.log.Info("model catalog loaded", "providers", len(catalog.Providers), "models", len(b.catalogModels))
models := applyModelCatalog(&b.cfg, catalog)
// applyModelCatalog writes into the boot struct, while every consumer
// reads config.Holder. Publishing the merged copy here is what puts the
// catalog's model limits and discovered providers into the running config;
// without it the first live update would be the first snapshot to carry
// them.
b.cfgHolder.Set(b.cfg.Clone())
b.setCatalogState(catalog, models)
b.log.Info("model catalog loaded", "providers", len(catalog.Providers), "models", len(models))
}
// seedSoul writes the starter SOUL beside the config on first run, so the
// identity a user can edit exists before the SOUL loader (#439) reads it. It
// runs on the full process's boot tail rather than in loadConfig: an offline
// command such as `-requeue` reads the config but must not write the operator's
// files. A failure is logged, not fatal -- the loader falls back to the embedded
// default, so a read-only config directory must not stop the process
// (docs/prds/soul.md, "Prompt and failure contract").
func (b *boot) seedSoul(cfgPath string) {
result, err := configuration.SeedSoul(cfgPath, agent.ShippedSoul())
if err != nil {
b.log.Warn("soul: starter file unavailable", "err", err)
return
}
if result.Action == configuration.SoulCreated {
b.log.Info("soul: starter file written", "path", result.Path, "action", result.Action)
}
}
// refreshModelCatalog re-reads the catalog and republishes everything derived
// from it, so a credential that becomes resolvable after boot or a model added
// upstream reaches the running config, the dashboard and the chat runtime
// without a restart. It is the live half of loadCatalog.
//
// A failed read keeps the snapshot already running: the model list must not go
// blank because the catalog service had a bad minute. The caller decides
// whether the failure is fatal to it -- the refresh loop logs and retries.
func (b *boot) refreshModelCatalog(ctx context.Context) error {
catalog, err := modelcatalog.Load(ctx, b.catalogOptions(b.catalogCachePath))
if err != nil {
return err
}
base := b.cfgHolder.Get().Clone()
models := applyModelCatalog(&base, catalog)
cfg, _, err := b.runtimeConfig(ctx, base)
if err != nil {
return err
}
// Committed only once the layered config passed validation, so a refused
// layering leaves the catalog the process is running alone.
b.setCatalogState(catalog, models)
b.publishConfig(ctx, cfg)
b.rebuildChatModelRuntime(cfg)
b.log.Info("model catalog refreshed", "providers", len(catalog.Providers), "models", len(models))
return nil
}
// catalogOptions is the one place the catalog read options are built: boot and
// every live refresh read the same endpoint, the same cache file and the same
// credential lookup. Getenv is the secrets registry, not the raw process
// environment, so a key a secret engine starts resolving after boot is visible
// to the next refresh. A variable exported into the shell after the process
// started is not, and cannot be: the process environment is fixed at exec.
func (b *boot) catalogOptions(cachePath string) modelcatalog.Options {
opts := modelcatalog.Options{
URL: b.catalogURL,
CachePath: cachePath,
}
if b.secrets != nil {
opts.Getenv = b.secrets.Getenv
}
if b.cfgHolder != nil {
opts.Configured = b.cfgHolder.Get().Providers
}
return opts
}
// modelCatalogRefreshInterval is how often the running daemon re-reads the
// model catalog. The document itself changes on the order of weeks, but a
// credential a secret engine starts resolving can appear at any moment, so the
// interval trades one cheap HTTP read against how long an operator waits for a
// newly usable provider to show up. It is deliberately the daemon's own
// cadence and not an operator setting: the catalog is derived state, not a
// stored resource, and a knob with no second consumer would be a setting that
// parses and does nothing.
const modelCatalogRefreshInterval = time.Hour
// startModelCatalogRefresh re-reads the catalog on an interval for the life of
// the process. It belongs to the daemon, which owns the published
// configuration; the standalone Gateway reads the catalog once at boot.
func (b *boot) startModelCatalogRefresh(ctx context.Context) {
every := b.catalogRefreshInterval
if every <= 0 {
every = modelCatalogRefreshInterval
}
go func() {
ticker := time.NewTicker(every)
defer ticker.Stop()
for {
select {
case <-ctx.Done():
return
case <-ticker.C:
if err := b.refreshModelCatalog(ctx); err != nil {
b.log.Warn("model catalog refresh failed; keeping the loaded catalog", "err", err)
}
}
}
}()
}
// setupObservability builds the event bus and dashboard server. Every event is persisted (stamped with its row id) and
@@ -698,7 +869,7 @@ func (b *boot) startEmbeddedNATS(ctx context.Context) (string, string, error) {
// not require Docker; repository tasks park instead of falling back to a host
// model loop.
func (b *boot) setupContainers(ctx context.Context) func() {
containerPool, storeBackend, closeDocker := startContainers(ctx, b.cfg, b.log)
containerPool, storeBackend, closeDocker := startContainers(ctx, b.cfgHolder, b.secrets, b.log)
b.containerPool = containerPool
b.storeBackend = storeBackend
return closeDocker
@@ -721,6 +892,26 @@ func (b *boot) setupEmbeddings(cfg config.Config, log *slog.Logger) {
}
}
// setupTranscriber builds the optional voice-transcription capability. It is
// the model-owning side of the Messaging Service boundary: the channel
// frontend carries a voice note's audio bytes across the inbound wire but
// holds no provider credential, so this process turns them into text before
// the turn's message is recorded. Like setupEmbeddings, a client is wired only
// when models["transcription"] names a provider/model and that provider's
// credential resolves; a configured-but-unusable role is logged so the
// degradation is visible, while a role never configured is silent.
func (b *boot) setupTranscriber(cfg config.Config, log *slog.Logger) {
client, ok := transcription.New(cfg.Models, cfg.Providers, transcription.Options{
ResolveSecret: b.secrets.Resolve,
})
if ok {
b.transcriber = client
log.Info("voice transcription enabled", "role", transcription.Role)
} else if cfg.Models[transcription.Role] != "" {
log.Warn("voice transcription configured but unavailable; capability disabled", "role", transcription.Role)
}
}
// setupLLMAndChat wires the runtime, tool registry, model management,
// personas and the dashboard's chat service.
func (b *boot) setupLLMAndChat(ctx context.Context) error {
@@ -1208,26 +1399,20 @@ func (b *boot) setupCurators(ctx context.Context) {
b.addCleanup(shutdownCuratorRegistry(reg, log))
}
func (b *boot) setupGuardrails() {
log := b.log
gc := tools.DefaultGuardrailConfig()
b.guardrails = tools.NewGuardrailEngine(gc)
log.Info(
"guardrail engine enabled",
"exact_failure_warn", gc.ExactFailureWarnAfter,
"same_tool_failure_warn", gc.SameToolFailureWarnAfter,
"no_progress_warn", gc.NoProgressWarnAfter,
)
}
// registerTools registers the tool providers with a lifecycle: workspace
// file/shell tools and optional MCP servers. Memory tools
// (memory_create/update/delete/list) are not registered here -- they are
// built per turn onto the resolved Subject's scopes
// (internal/gateway/turn_memory_tool.go), not once at boot.
func (b *boot) registerTools(ctx context.Context) error {
cfg, log := b.cfg, b.log
// The running config, not b.cfg: boot is where the first stored version of
// every kind was already layered in, and a watch that applied a later one
// before this point published through the holder.
cfg, log := b.cfgHolder.Get(), b.log
b.providerRegistry = toolprovider.NewRegistry(b.toolReg)
b.mcpMu.Lock()
b.mcpApplied = make(map[string]appliedMCPServer, len(cfg.Tools.MCPServers))
b.mcpMu.Unlock()
// Workspace file and shell tools. Registered only when a workspace is
// configured: these read, write and execute, so the directory is a
// deliberate choice rather than a default.
@@ -1246,7 +1431,7 @@ func (b *boot) registerTools(ctx context.Context) error {
log.Info("workspace tools disabled (chat.workspace is unset)")
}
for _, srv := range cfg.Tools.MCPServers {
provider, err := configuredMCPProvider(srv, cfg.WorkDir)
provider, err := b.buildMCPProvider(srv)
if err != nil {
log.Warn("mcp tool provider skipped", "name", srv.Name, "err", err)
continue
@@ -1261,6 +1446,9 @@ func (b *boot) registerTools(ctx context.Context) error {
log.Warn("mcp tool provider skipped", "name", srv.Name, "err", err)
continue
}
b.mcpMu.Lock()
b.mcpApplied[strings.TrimSpace(srv.Name)] = appliedMCPServer{server: srv, id: provider.Manifest().ID}
b.mcpMu.Unlock()
}
if err := b.capabilityHost.Register(b.providerRegistry); err != nil {
log.Error("tool-provider capability registration failed", "err", err)
@@ -1310,21 +1498,7 @@ func (b *boot) registerStandaloneTools() {
// web_fetch. Registered directly rather than as a tool provider: it has
// no process to start or stop, so the provider lifecycle would buy
// nothing. Disabled by configuration returns nil and advertises nothing.
if entry := webfetch.Tool(webfetch.Config{
Enabled: cfg.Tools.WebFetch.IsEnabled(),
Timeout: cfg.Tools.WebFetch.Timeout.Std(),
MaxBytes: cfg.Tools.WebFetch.MaxBytes,
AllowPrivateNetworks: cfg.Tools.WebFetch.AllowPrivateNetworks,
}); entry != nil {
if err := b.toolReg.Register(*entry); err != nil {
log.Warn("web_fetch registration failed", "err", err)
} else {
log.Info("web fetch enabled",
"allow_private_networks", cfg.Tools.WebFetch.AllowPrivateNetworks)
}
} else {
log.Info("web fetch disabled")
}
b.registerWebFetchTool(cfg)
// send_file. Rooted at the same workspace as the file tools and gated
// by the same confinement, because it hands a host file to an outbound
@@ -1343,6 +1517,30 @@ func (b *boot) registerStandaloneTools() {
b.registerMinimaxTool(cfg, log)
}
// registerWebFetchTool registers web_fetch from cfg. Split out of
// registerStandaloneTools because a live tool-settings change rebuilds the
// entry: the one a boot built captured its config at construction, so it
// cannot read a changed value per call.
func (b *boot) registerWebFetchTool(cfg config.Config) {
log := b.log
entry := webfetch.Tool(webfetch.Config{
Enabled: cfg.Tools.WebFetch.IsEnabled(),
Timeout: cfg.Tools.WebFetch.Timeout.Std(),
MaxBytes: cfg.Tools.WebFetch.MaxBytes,
AllowPrivateNetworks: cfg.Tools.WebFetch.AllowPrivateNetworks,
})
if entry == nil {
log.Info("web fetch disabled")
return
}
if err := b.toolReg.Register(*entry); err != nil {
log.Warn("web_fetch registration failed", "err", err)
return
}
log.Info("web fetch enabled",
"allow_private_networks", cfg.Tools.WebFetch.AllowPrivateNetworks)
}
// registerMinimaxTool registers generate_video. Off by default -- see
// MinimaxConfig's doc comment, it spends real API credits per call -- and,
// unlike web_fetch, needs a resolved credential before it can be
@@ -1390,7 +1588,6 @@ func (b *boot) buildDaemon() {
Bus: b.bus,
Forge: b.forgeClient,
Trees: b.trees,
CapabilityHost: b.capabilityHost,
Storage: b.storeBackend,
Log: log,
Tasks: b.natsClient,
@@ -1399,8 +1596,6 @@ func (b *boot) buildDaemon() {
StateStoreGrants: b.stateStoreGrants,
ContainerPool: b.containerPool,
KitLauncher: b.kitLauncher,
Guardrails: b.guardrails,
ToolRegistry: b.toolReg,
Identities: b.identityRunners,
RootIdentityID: identity.StableID(configuredIdentityNames(b.cfg)[0]),
TaskLogs: b.taskLogs,
@@ -1504,15 +1699,9 @@ func (b *boot) setupForgeWebhook() {
// The reaction publisher is the consumer's supply side: the same typed
// reaction the poll produces, keyed source-independently, so webhook and
// poll deliveries of one review dedup (pr-review-remediation.md
// decision 2).
publishReaction := func(ctx context.Context, reaction workintake.ReviewCommentEnvelope) error {
payload, err := reaction.Encode()
if err != nil {
return err
}
return b.d.Tasks.PublishUnique(ctx, reaction.Subject(), reaction.IdempotencyKey(), payload)
}
receiver := forgewebhook.New(secretValue, cfg.Dispatch.Trigger, cfg.Label, cfg.BotUser, b.d.PublishTask, publishReaction, log)
// decision 2). The daemon's PublishReaction is the single reaction write
// path: it stamps the root org a webhook delivery would otherwise lack.
receiver := forgewebhook.New(secretValue, cfg.Dispatch.Trigger, cfg.Label, cfg.BotUser, b.d.PublishTask, b.d.PublishReaction, log)
host, port := parseListenAddr(cfg.Forge.WebhookAddr, "0.0.0.0", 8645)
addr := fmt.Sprintf("%s:%d", host, port)
srv := &http.Server{Addr: addr, Handler: receiver}
@@ -1562,10 +1751,11 @@ func (b *boot) configOrigins() []webui.ConfigOrigin {
// the daemon's own configuration state. The daemon is the configuration
// owner, so it is the process that renders this view (archie-core-ml30).
func (b *boot) configViewInput(ctx context.Context) webui.ConfigViewInput {
catalog, _ := b.catalogState()
in := webui.ConfigViewInput{
Config: b.cfgHolder.Get(),
Provenance: b.configOrigins(),
Catalog: catalogView(b.catalog),
Catalog: catalogView(catalog),
}
if b.lastReload != nil {
status := b.lastReload()
+1 -1
View File
@@ -109,7 +109,7 @@ func TestConnectEmbeddedNATSUsesManagedWorkerBridge(t *testing.T) {
DockerClient: dockerClient,
Network: "workers",
RequireHostGateway: true,
}, slog.New(slog.DiscardHandler))
}, nil, slog.New(slog.DiscardHandler))
if err != nil {
t.Fatalf("container.NewPool = %v", err)
}
+40 -14
View File
@@ -18,10 +18,13 @@ type chatModelManager struct {
mu sync.RWMutex
models []string
active string
// catalogs are the non-role sources the model set is derived from, kept
// so SetConfigured can rebuild it after a live settings update without a
// caller re-passing them.
catalogs [][]string
// configured, explicit and catalog are the sources the offered set is
// derived from: the role assignments (live via SetConfigured), the
// [chat].models catalog (fixed for the process), and the model catalog's
// own refs (replaced wholesale by SetModelCatalog).
configured map[string]string
explicit []string
catalog []string
details map[string]gateway.ModelDetails
providerNames map[string]string
}
@@ -63,13 +66,29 @@ func mergeModelRefs(configured map[string]string, catalogs ...[]string) []string
}
func newChatModelManager(configured map[string]string, catalogs ...[]string) *chatModelManager {
return &chatModelManager{
models: mergeModelRefs(configured, catalogs...),
active: pickChatDefault(configured),
catalogs: append([][]string(nil), catalogs...),
explicit := make([]string, 0, len(catalogs))
for _, catalog := range catalogs {
explicit = append(explicit, catalog...)
}
m := &chatModelManager{
configured: configured,
explicit: explicit,
details: make(map[string]gateway.ModelDetails),
providerNames: make(map[string]string),
}
m.active = pickChatDefault(configured)
m.setModelsLocked()
return m
}
// setModelsLocked re-derives the offered set from the three sources and keeps
// the active selection when the new set still carries it, falling back the way
// construction does when it does not.
func (m *chatModelManager) setModelsLocked() {
m.models = mergeModelRefs(m.configured, m.explicit, m.catalog)
if !slices.Contains(m.models, m.active) {
m.active = pickChatDefault(m.configured)
}
}
// SetConfigured re-derives the model set the manager offers after a live
@@ -81,16 +100,22 @@ func newChatModelManager(configured map[string]string, catalogs ...[]string) *ch
func (m *chatModelManager) SetConfigured(configured map[string]string) {
m.mu.Lock()
defer m.mu.Unlock()
models := mergeModelRefs(configured, m.catalogs...)
if !slices.Contains(models, m.active) {
m.active = pickChatDefault(configured)
}
m.models = models
m.configured = configured
m.setModelsLocked()
}
func (m *chatModelManager) ApplyModelCatalog(snapshot modelcatalog.Snapshot) {
// SetModelCatalog replaces the model catalog's contribution to the manager:
// the references it offers, the provider display names and the per-model
// details. The previous snapshot's entries are dropped rather than merged --
// a model or provider the catalog no longer publishes must stop being offered
// and described, and these maps are the manager's own state rather than a
// second catalogue to fall back on.
func (m *chatModelManager) SetModelCatalog(snapshot modelcatalog.Snapshot, models []string) {
m.mu.Lock()
defer m.mu.Unlock()
m.catalog = slices.Clone(models)
m.details = make(map[string]gateway.ModelDetails)
m.providerNames = make(map[string]string)
for _, provider := range snapshot.Providers {
m.providerNames[provider.ID] = provider.Name
for _, model := range provider.Models {
@@ -104,6 +129,7 @@ func (m *chatModelManager) ApplyModelCatalog(snapshot modelcatalog.Snapshot) {
}
}
}
m.setModelsLocked()
}
func (m *chatModelManager) ProviderDisplayName(provider string) string {
+14 -1
View File
@@ -12,8 +12,10 @@ import (
"github.com/samcharles93/ai-sdk/core"
"github.com/samcharles93/ai-sdk/runtime"
"github.com/samcharles93/archie-core/internal/agentexec"
"github.com/samcharles93/archie-core/internal/config"
"github.com/samcharles93/archie-core/internal/daemon"
"github.com/samcharles93/archie-core/internal/domain/messaging"
"github.com/samcharles93/archie-core/internal/events"
"github.com/samcharles93/archie-core/internal/gateway"
"github.com/samcharles93/archie-core/internal/installtype"
@@ -29,6 +31,10 @@ type chatSetup struct {
// running yet at this point, so the reload-safe Holder is mostly a
// formality, but matching the daemon's API keeps callers honest.
Cfg *config.Holder
// ToolLimits resolves the tool-output policy at each turn, so a live
// tool-settings update reaches the next chat turn's tool options rather
// than the value captured when the runner was built.
ToolLimits func() agentexec.ToolLimits
// LLM resolves the provider runtime at each use: a live provider-settings
// or model-role-assignments update swaps the runtime wholesale (ai-sdk's
// Runtime caches the provider instances it built), so the pointer cannot
@@ -63,6 +69,12 @@ type chatSetup struct {
// process makes, which the /status health source reads back. Nil disables
// recording (the recorder's methods tolerate that rather than panicking).
ProviderOutcomes *providerOutcomeRecorder
// Transcriber is the voice-transcription capability built by
// setupChatRuntime from this process's own [models]/[providers]. Every turn
// runner carries it so a speech attachment is transcribed before its
// message is recorded. Nil disables the capability (no role configured, or
// the provider credential did not resolve), leaving the frontend's note.
Transcriber messaging.Transcriber
}
// daemonRunningVersions reports the component versions this process can
@@ -175,7 +187,7 @@ func newChatTurnRunner(
Sessions: sessionStore,
Models: s.ChatModels,
Personas: s.Personas,
Model: newChatTurnModel(s.LLM, s.ToolReg, cfg.Chat.MaxSteps, toolLimits(cfg), s.ProviderOutcomes),
Model: newChatTurnModel(s.LLM, s.ToolReg, cfg.Chat.MaxSteps, s.ToolLimits, s.ProviderOutcomes),
TaskLister: s.ChatTaskLister,
Tasks: s.ChatTasks,
TaskLogs: s.ChatTaskLogs,
@@ -190,6 +202,7 @@ func newChatTurnRunner(
MemoryEngine: s.MemoryEngine,
MemoryWriter: s.MemoryWriter,
UserIdentity: userIdentityResolver(),
Transcriber: s.Transcriber,
Log: s.Log,
})
if err := runner.Recover(ctx); err != nil && s.Log != nil {
+33 -7
View File
@@ -23,7 +23,10 @@ type chatTurnModel struct {
llm func() *runtime.Runtime
registry *tools.Registry
maxSteps int
limits agentexec.ToolLimits
// limits resolves the tool-output policy at each turn, so a live
// tool-settings update is the policy the next turn builds its tool options
// with; a snapshot taken at construction would keep the boot limits.
limits func() agentexec.ToolLimits
// outcomes records each call's result for /status (see sendChatTurn).
outcomes *providerOutcomeRecorder
}
@@ -32,7 +35,7 @@ func newChatTurnModel(
llm func() *runtime.Runtime,
registry *tools.Registry,
maxSteps int,
limits agentexec.ToolLimits,
limits func() agentexec.ToolLimits,
outcomes *providerOutcomeRecorder,
) gateway.TurnModel {
return &chatTurnModel{
@@ -48,7 +51,11 @@ func (m *chatTurnModel) Prepare(
ctx context.Context,
req gateway.TurnPrepareContext,
) (gateway.PreparedTurnModel, error) {
options, err := chatGenerateOptions(ctx, nil, m.registry, m.maxSteps, m.limits, req.Extra, req.ContextWindow)
limits := agentexec.ToolLimits{}
if m.limits != nil {
limits = m.limits()
}
options, err := chatGenerateOptions(ctx, nil, m.registry, m.maxSteps, limits, req.Extra, req.ContextWindow)
if err != nil {
return nil, err
}
@@ -60,6 +67,7 @@ func (m *chatTurnModel) Prepare(
llm: m.llm(),
model: req.Model,
options: options,
reasoning: req.Reasoning,
toolInfo: toolSummaries(options.Tools),
toolTokens: gateway.EstimateTokens(string(toolSchema)),
outcomes: m.outcomes,
@@ -68,9 +76,13 @@ func (m *chatTurnModel) Prepare(
}
type preparedChatTurnModel struct {
llm *runtime.Runtime
model string
options core.GenerateOptions
llm *runtime.Runtime
model string
options core.GenerateOptions
// reasoning is the catalog's class for this model, carried from the
// prepare context so Generate can choose the output-token parameter the
// provider accepts.
reasoning bool
toolInfo []gateway.ToolSummary
toolTokens int
outcomes *providerOutcomeRecorder
@@ -178,6 +190,20 @@ func (m *preparedChatTurnModel) ToolSchemaTokens() int {
return m.toolTokens
}
// maxTokensForRequest returns the output-token bound to place on an ai-sdk
// generate request for a model. A reasoning-class model gets no bound at all:
// the OpenAI chat-completions provider serialises a non-zero bound as
// `max_tokens`, which reasoning models reject with HTTP 400 ("Use
// 'max_completion_tokens' instead"), and the pinned ai-sdk release has no
// max_completion_tokens path, so no bound is the only usable form. A classic
// model keeps the caller's bound unchanged.
func maxTokensForRequest(reasoning bool, bound int) int {
if reasoning {
return 0
}
return bound
}
func (m *preparedChatTurnModel) Generate(
ctx context.Context,
request gateway.TurnModelRequest,
@@ -187,7 +213,7 @@ func (m *preparedChatTurnModel) Generate(
options.Messages = buildTurnMessages(request)
// This is one provider response's output allowance, not a turn-
// continuation budget. Tool loops remain free to continue.
options.MaxTokens = request.MaxOutputTokens
options.MaxTokens = maxTokensForRequest(m.reasoning, request.MaxOutputTokens)
// A runtime swapped to nil -- a live update that left no usable provider
// configured -- is a refused turn, not a panic.
if m.llm == nil {
+134 -7
View File
@@ -2,12 +2,14 @@ package archied
import (
"context"
"encoding/json"
"fmt"
"time"
"github.com/samcharles93/archie-core/internal/agentexec"
"github.com/samcharles93/archie-core/internal/app/controlplane"
"github.com/samcharles93/archie-core/internal/config"
"github.com/samcharles93/archie-core/internal/domain/applystatus"
"github.com/samcharles93/archie-core/internal/domain/workflow"
"github.com/samcharles93/archie-core/internal/infrastructure/configuration"
)
@@ -75,7 +77,8 @@ func (b *boot) reloadConfig(ctx context.Context, doc *configuration.Document) er
// drop them from the running config even though the file is
// unchanged. Re-apply the same merge (idempotent: a catalog that
// failed to load merges to identity).
applyModelCatalog(&doc.Config, b.catalog)
catalog, _ := b.catalogState()
applyModelCatalog(&doc.Config, catalog)
// Bounded: this runs on the signal loop, which handles nothing else
// while it waits, and a State Store that has stopped answering must
// surface as a failed reload rather than a SIGHUP that never returns.
@@ -89,6 +92,10 @@ func (b *boot) reloadConfig(ctx context.Context, doc *configuration.Document) er
old := b.cfgHolder.Get()
b.currentProvenance.Store(&doc.Provenance)
b.publishConfig(ctx, cfg)
// The container pool reads the published config on every acquire, but the
// dispatcher holds its limit; a reloaded containers.max_concurrency must
// reach it the same way a control-plane update does.
b.resizeTaskDispatcher(cfg.Containers.MaxConcurrency)
if fields := changedNonReloadableFields(old, cfg); len(fields) > 0 {
b.log.Warn("config reloaded; some changes require a restart",
"fields", fields, "paths", doc.Provenance.Paths())
@@ -257,16 +264,34 @@ func (b *boot) startLiveSettings(ctx context.Context) error {
// (b.runtimeConfig, which applies the execution budgets the settings watch
// published) and republishes through config.Holder, so the change takes
// effect without a restart. Kinds stay out of this list while a
// startup-built component still holds their value -- tool, plugin, container
// and channel settings are frozen in components built at boot and remain
// restart-required -- and a kind joins it only with a consumer that re-reads
// it. Each kind is watched in its own goroutine, exactly the shape the
// startup-built component still holds their value; none does today, so the
// list is every control-plane kind. A kind joins it only with a consumer that
// re-reads it. ToolSettingsKind is on it because applyRuntimeResourceUpdate
// reconciles the MCP provider set and rebuilds the two config-derived tool
// entries (boot.reconcileToolSettings, its consumer); a removed MCP server
// really disconnects, because it is a child process with a Stop, not a Yaegi
// interpreter. ContainerRuntimePoliciesKind is on it because both consumers
// re-read the published config: the container pool takes its image, pull
// policy, network, max-uptime and concurrency cap per acquire, and the daemon's
// dispatcher is resized after the publish (applyRuntimeResourceUpdate below).
// PluginSettingsKind is on the list because its only consumer is the directory
// reconciliation, which reads the running config each tick; a removed plugin
// or engine still cannot unload, so that kind reports the removal rather than
// requiring a restart. ChannelSettingsKind is on the list for this process's
// own re-layer; the Messaging Service, which owns the channel transports,
// reconciles them separately and restarts only the channel whose settings
// changed. Each kind is watched in its own goroutine, exactly the shape the
// workflow-execution-settings watch established.
var runtimeResourceKinds = []string{
controlplane.ProviderSettingsKind,
controlplane.ModelRoleAssignmentsKind,
controlplane.RepositoryPoliciesKind,
controlplane.SchedulingPolicyKind,
controlplane.ToolSettingsKind,
controlplane.PluginSettingsKind,
controlplane.ReviewSettingsKind,
controlplane.ChannelSettingsKind,
controlplane.ContainerRuntimePoliciesKind,
}
// startRuntimeResourceWatches keeps a watch per live kind established for the
@@ -319,12 +344,20 @@ func (b *boot) applyRuntimeResourceUpdate(ctx context.Context, kind string, upda
b.log.Error("runtime settings watch failed", "kind", kind, "err", update.Err)
return
}
if kind == controlplane.PluginSettingsKind {
if err := b.refuseSkillsDirChange(ctx); err != nil {
b.applyStatus.Report(ctx, kind, update.Version, err)
b.log.Error("plugin settings refused; the running settings stay", "version", update.Version, "err", err)
return
}
}
base := b.cfgHolder.Get()
// Boot publishes before the model catalog is merged in, so the snapshot
// this read returns can lack the catalog's model limits and discovered
// providers. The merge is idempotent, and reloadConfig re-applies it
// before layering for the same reason.
applyModelCatalog(&base, b.catalog)
catalog, _ := b.catalogState()
applyModelCatalog(&base, catalog)
cfg, _, err := b.runtimeConfig(ctx, base)
if err != nil {
// runtimeConfig reported the refusal through apply status for every
@@ -337,19 +370,113 @@ func (b *boot) applyRuntimeResourceUpdate(ctx context.Context, kind string, upda
switch kind {
case controlplane.ProviderSettingsKind, controlplane.ModelRoleAssignmentsKind:
b.rebuildChatModelRuntime(cfg)
case controlplane.ToolSettingsKind:
// The document's server set and tool entries are live now; reconcile
// them and report the reconciliation's own outcome, which is the one
// that knows whether a server failed to connect.
if err := b.reconcileToolSettings(ctx, cfg); err != nil {
b.applyStatus.Report(ctx, kind, update.Version, err)
b.log.Error("tool settings reconciled with errors; the running servers stay", "version", update.Version, "err", err)
}
case controlplane.PluginSettingsKind:
// The document's directories are live now; load what they hold and
// report the reconciliation's own outcome, which is the one that knows
// whether a removal is outstanding.
b.reconcileRuntimePlugins(ctx)
case controlplane.ContainerRuntimePoliciesKind:
// The pool reads the republished config itself; the dispatcher holds
// its limit, so a raise must be woken to admit queued work now and a
// lower must be honoured as running tasks finish.
b.resizeTaskDispatcher(cfg.Containers.MaxConcurrency)
}
}
// resizeTaskDispatcher applies a new containers.max_concurrency to the
// daemon's running dispatcher. The pool needs no poke because it reads the
// published config on every acquire. b.d is nil in the boot window in which
// the watches are already live but the daemon is not built yet; there the
// dispatcher is created later from the config the next drain reads.
func (b *boot) resizeTaskDispatcher(maxConcurrency int) {
if b.d != nil {
b.d.ResizeTaskDispatcher(maxConcurrency)
}
}
// startPluginReconcile establishes the plugin, module and secret-engine
// directory reconciliation on the apply-status restamp interval. Boot calls it
// after loadPlugins, loadWorkflows and configuredSecretRegistry have loaded
// the directories, so the seed records what is already running and the first
// tick loads only what appears or changes afterwards.
func (b *boot) startPluginReconcile(ctx context.Context) {
r := newPluginReconciler(b.log, pluginReconcileTargets{
host: b.capabilityHost,
secrets: b.secrets,
modules: b.modules,
dirs: func() config.Config { return b.cfgHolder.Get() },
relayer: func(ctx context.Context) error {
cfg, _, err := b.runtimeConfig(ctx, b.cfgHolder.Get())
if err != nil {
return err
}
b.publishConfig(ctx, cfg)
return nil
},
report: func(ctx context.Context, err error) {
b.applyStatus.Report(ctx, controlplane.PluginSettingsKind, b.applyStatus.AppliedVersion(controlplane.PluginSettingsKind), err)
},
})
b.pluginReconciler = r
r.seed()
go r.run(ctx, applystatus.RestampInterval)
}
// reconcileRuntimePlugins runs one directory reconciliation after a stored
// plugin-settings change, so a new directory loads on the tick the document
// lands rather than waiting for the poller.
func (b *boot) reconcileRuntimePlugins(ctx context.Context) {
if b.pluginReconciler == nil {
return
}
_ = b.pluginReconciler.reconcile(ctx)
}
// refuseSkillsDirChange refuses a stored plugin-settings document whose
// skills_dir differs from the value in force. skills_dir is the one directory
// with no live consumer, so a change to it stays a restart-scoped edit rather
// than a value the daemon reports as applied. The check reads the running
// config, so it cannot live in validatePluginSettings, which is value-agnostic
// by the archie-core-1143 decision.
func (b *boot) refuseSkillsDirChange(ctx context.Context) error {
var stored struct {
SkillsDir string `json:"skills_dir"`
}
_, found, err := b.controlPlane.Query(ctx, controlplane.PluginSettingsKind, func(value []byte) error {
return json.Unmarshal(value, &stored)
})
if err != nil {
return fmt.Errorf("read plugin settings: %w", err)
}
if !found {
return nil
}
if running := b.cfgHolder.Get().SkillsDir; stored.SkillsDir != running {
return fmt.Errorf("skills_dir applies on restart: stored %q, running %q", stored.SkillsDir, running)
}
return nil
}
// rebuildChatModelRuntime re-derives the gateway chat runtime's provider set
// after a live change to provider-settings or model-role-assignments. The
// turn runner reads the runtime through boot.chatLLM, so the swap reaches its
// next turn without rebuilding the runner, and the model manager re-derives
// the references it offers from the new role assignments.
// the references it offers from the new role assignments and the catalog.
func (b *boot) rebuildChatModelRuntime(cfg config.Config) {
if b.chatModels == nil {
return // this process serves no chat turns
}
b.setLLM(agentexec.NewRuntime(executionProviders(cfg)))
b.chatModels.SetConfigured(cfg.Models)
catalog, models := b.catalogState()
b.chatModels.SetModelCatalog(catalog, models)
b.log.Info("chat model runtime rebuilt", "providers", len(cfg.Providers), "models", len(cfg.Models))
}
@@ -15,6 +15,7 @@ import (
"github.com/samcharles93/archie-core/internal/agentexec"
"github.com/samcharles93/archie-core/internal/app/controlplane"
pb "github.com/samcharles93/archie-core/internal/contracts/controlplane/v1"
"github.com/samcharles93/archie-core/internal/daemon"
"github.com/samcharles93/archie-core/internal/secret"
)
@@ -240,3 +241,33 @@ func TestLiveModelRoleUpdateRebuildsTheChatModelRuntime(t *testing.T) {
})
}
}
// TestLiveContainerRuntimePoliciesReLayerAndResizeTheDispatcher pins the
// container kind's live seam (archie-core-zfb0.2) at the apply path: a stored
// container-runtime-policies document re-layers into the running config the
// pool reads per acquire, and the daemon resizes its running dispatcher from
// the same publish rather than requiring a restart.
func TestLiveContainerRuntimePoliciesReLayerAndResizeTheDispatcher(t *testing.T) {
values := databaseOwnedResources()
values[controlplane.ContainerRuntimePoliciesKind] = map[string]any{
"image": "archie:live", "max_concurrency": 5, "pull_policy": "missing",
}
stub := &resourceWatchStub{values: values, version: 3, storedVersion: map[string]int64{}, opens: map[string]int{}}
b, _ := newResourceWatchBoot(t, stub)
if err := b.loadRuntimeConfig(t.Context()); err != nil {
t.Fatalf("loadRuntimeConfig: %v", err)
}
// A running daemon, so the resize arm executes against a real dispatcher
// rather than taking the boot-window nil guard.
b.d = &daemon.Daemon{Cfg: b.cfgHolder, Log: b.log}
b.applyRuntimeResourceUpdate(t.Context(), controlplane.ContainerRuntimePoliciesKind, controlplane.AppliedResource{Version: 4})
got := b.cfgHolder.Get().Containers
if got.MaxConcurrency != 5 {
t.Errorf("running max_concurrency = %d, want the stored 5", got.MaxConcurrency)
}
if got.Image != "archie:live" {
t.Errorf("running image = %q, want the stored archie:live", got.Image)
}
}
+16 -1
View File
@@ -12,6 +12,7 @@ import (
"github.com/samcharles93/archie-core/internal/config"
"github.com/samcharles93/archie-core/internal/domain/applystatus"
"github.com/samcharles93/archie-core/internal/domain/messaging"
"github.com/samcharles93/archie-core/internal/events"
"github.com/samcharles93/archie-core/internal/gateway"
"github.com/samcharles93/archie-core/internal/infrastructure/gatewayrpc"
@@ -71,6 +72,7 @@ func RunGateway(ctx context.Context, options GatewayOptions) error {
return err
}
b.loadCatalog(ctx, options.Config)
b.seedSoul(options.Config)
url, token := b.cfg.NATS.URL, ""
var nc *natsio.Conn
var err error
@@ -189,18 +191,30 @@ func (b *boot) startGatewayRuntime(ctx context.Context, actor gateway.ChatTaskAc
return contract, nil
}
// inboundMessageHeadroomBytes leaves room for the protobuf framing and the
// rest of the request around an attachment at the messaging contract's
// ceiling, so the largest attachment a channel frontend may send is accepted
// rather than rejected as an oversized gRPC message.
const inboundMessageHeadroomBytes = 1 << 20
// gatewayServerOpts applies the transport security boundary: a loopback
// listener is served insecure with no token; any non-loopback address
// requires a Bearer [REDACTED], else the process fails closed. It mirrors
// stateStoreServerOpts, minus the State Store's task-scoped grants: the
// Gateway has one administrative token for all callers.
//
// The receive limit is raised above gRPC's 4 MB default because an inbound
// request carries a channel attachment's bytes: the channel frontend
// downloads the file, and this process -- which runs the turn -- holds no
// platform credential to fetch it again.
func gatewayServerOpts(listen, token string) (opts []grpc.ServerOption, loopback bool, err error) {
recv := grpc.MaxRecvMsgSize(messaging.MaxInboundAttachmentBytes + inboundMessageHeadroomBytes)
loopback, err = gatewayrpc.TargetIsLoopback(listen)
if err != nil {
return nil, false, err
}
if loopback {
return nil, true, nil
return []grpc.ServerOption{recv}, true, nil
}
if token == "" {
return nil, false, fmt.Errorf(
@@ -209,6 +223,7 @@ func gatewayServerOpts(listen, token string) (opts []grpc.ServerOption, loopback
)
}
return []grpc.ServerOption{
recv,
grpc.ChainUnaryInterceptor(gatewayrpc.UnaryServerInterceptor(token)),
grpc.ChainStreamInterceptor(gatewayrpc.StreamServerInterceptor(token)),
}, false, nil
+27 -5
View File
@@ -9,7 +9,9 @@ import (
"github.com/samcharles93/archie-core/internal/config"
"github.com/samcharles93/archie-core/internal/domain/agent"
"github.com/samcharles93/archie-core/internal/domain/storecontract"
"github.com/samcharles93/archie-core/internal/domain/taskactions"
"github.com/samcharles93/archie-core/internal/gateway"
"github.com/samcharles93/archie-core/internal/taskstate"
"github.com/samcharles93/archie-core/internal/tools"
)
@@ -23,9 +25,15 @@ func (b *boot) setupChatRuntime(ctx context.Context, cfg config.Config) error {
// non-command message processing.
providers := executionProviders(cfg)
b.setLLM(agentexec.NewRuntime(providers))
// Transcription is a model-role capability, like the chat models it sits
// beside: it is built here, on the model-owning side, from this process's
// own [models]/[providers]. The Messaging Service carries a voice note's
// bytes to the Gateway and never holds the credential.
b.setupTranscriber(cfg, b.log)
b.toolReg = tools.NewRegistry()
chatModels := newChatModelManager(cfg.Models, cfg.Chat.Models, b.catalogModels)
chatModels.ApplyModelCatalog(b.catalog)
chatModels := newChatModelManager(cfg.Models, cfg.Chat.Models)
catalog, catalogModels := b.catalogState()
chatModels.SetModelCatalog(catalog, catalogModels)
b.chatModels = chatModels
// ── Persona registry ─────────────────────────────────────────────
@@ -63,8 +71,14 @@ func (b *boot) setupChatRuntime(ctx context.Context, cfg config.Config) error {
canceller = ec
}
b.chatController = gateway.NewStoreTaskController(chatTaskControllerAdapter{
taskByID: b.stateStore.TaskByID,
requeue: b.stateStore.Requeue,
taskByID: b.stateStore.TaskByID,
// Chat's /approve reaches the daemon's one task-action service rather
// than its own requeue, so a chat approval and a dashboard approval
// cannot record different decisions for one operator intent
// (docs/prds/pr-review-operator-response.md, Decision 1).
approve: func(ctx context.Context, scope *string, actor taskactions.Actor, taskID int64, res taskactions.ActionPayload) error {
return b.taskActions().Apply(ctx, scope, actor, taskID, taskstate.ActionApprove, res)
},
cancelExecution: canceller.CancelExecution,
})
b.updateService = makeUpdateService(chatSetup{Cfg: config.NewHolder(cfg)})
@@ -123,8 +137,13 @@ func (b *boot) setupGatewayChat(ctx context.Context, actor gateway.ChatTaskActor
router.InitSessions(b.chatSessionStore)
configureTaskCommands(router, b.chatTasks, b.chatController, chatTaskListerAdapter{tasks: b.stateStore.Tasks}, b.defaultChatIdentity)
router.Health = b.statusHealth
// ToolLimits reads the live holder, not cfg: the daemon's runtime-resource
// watch republishes a tool-settings change through it, so a policy edit
// applies to the next turn in the process that serves this runner.
setup := chatSetup{
Cfg: config.NewHolder(cfg), LLM: b.chatLLM, ChatModels: b.chatModels, ToolReg: b.toolReg,
Cfg: config.NewHolder(cfg),
ToolLimits: func() agentexec.ToolLimits { return toolLimits(b.cfgHolder.Get()) },
LLM: b.chatLLM, ChatModels: b.chatModels, ToolReg: b.toolReg,
Personas: b.personas, ChatTasks: b.chatTasks,
ChatTaskLister: chatTaskListerAdapter{tasks: b.stateStore.Tasks},
ChatTaskLogs: chatTaskLogReaderAdapter{tasks: b.stateStore.TaskByID, taskLogs: b.taskLogs},
@@ -133,6 +152,9 @@ func (b *boot) setupGatewayChat(ctx context.Context, actor gateway.ChatTaskActor
Bus: b.bus, Log: b.log,
MemoryEngine: b.memoryStore(),
MemoryWriter: b.memoryWriter(),
// The transcription capability is built beside the chat runtime, from
// the model-role configuration only this process reads.
Transcriber: b.transcriber,
// The Gateway executes the web chat's turns, so it is the process
// that must record their outcomes for /status.
ProviderOutcomes: b.providerOutcomes,
+81 -6
View File
@@ -2,8 +2,11 @@ package archied
import (
"context"
"errors"
"io"
"log/slog"
"net"
"sync"
"testing"
"time"
@@ -25,8 +28,10 @@ func TestGatewayServerOptsLoopbackIsInsecure(t *testing.T) {
if !loopback {
t.Fatal("loopback listener should report loopback")
}
if len(opts) != 0 {
t.Fatalf("loopback listener should have no server opts, got %d", len(opts))
// Only the inbound size limit; a loopback listener installs no auth
// interceptors (the count is the only handle on opaque ServerOptions).
if len(opts) != 1 {
t.Fatalf("loopback listener should install only the inbound size limit, got %d options", len(opts))
}
}
@@ -41,8 +46,8 @@ func TestGatewayServerOptsNonLoopbackRequiresToken(t *testing.T) {
if loopback {
t.Fatal("non-loopback listener should not report loopback")
}
if len(opts) != 2 {
t.Fatalf("non-loopback listener should install two server options (unary and stream auth interceptors), got %d", len(opts))
if len(opts) != 3 {
t.Fatalf("non-loopback listener should install three server options (inbound size limit plus unary and stream auth interceptors), got %d", len(opts))
}
}
@@ -89,11 +94,11 @@ func (stubGatewayContract) SetPersona(context.Context, string, string) (bool, er
return false, nil
}
func (stubGatewayContract) ApplyTaskAction(context.Context, string, int64, taskstate.Action) (messaging.TaskActionResult, error) {
func (stubGatewayContract) ApplyTaskAction(context.Context, string, int64, taskstate.Action, taskactions.ActionPayload) (messaging.TaskActionResult, error) {
return messaging.TaskActionResult{}, nil
}
func (stubGatewayContract) ApplyOperatorTaskAction(context.Context, taskactions.Actor, int64, taskstate.Action) (messaging.TaskActionResult, error) {
func (stubGatewayContract) ApplyOperatorTaskAction(context.Context, taskactions.Actor, int64, taskstate.Action, taskactions.ActionPayload) (messaging.TaskActionResult, error) {
return messaging.TaskActionResult{}, nil
}
@@ -135,3 +140,73 @@ func TestGatewayServingAnnouncesReady(t *testing.T) {
t.Fatalf("serveGatewayListener returned: %v", err)
}
}
// recordingGatewayContract captures the inbound a Stream call reached the
// Gateway with, so a test can prove the request's payload survived the wire.
type recordingGatewayContract struct {
stubGatewayContract
mu sync.Mutex
last messaging.Inbound
}
func (c *recordingGatewayContract) Stream(_ context.Context, in messaging.Inbound) (<-chan messaging.ChatEvent, error) {
c.mu.Lock()
c.last = in
c.mu.Unlock()
events := make(chan messaging.ChatEvent)
close(events)
return events, nil
}
// TestGatewayAcceptsAnInboundAttachmentOverTheDefaultGRPCLimit: the inbound
// request now carries a channel attachment's bytes, and gRPC's default receive
// limit is 4 MB -- well under the messaging contract's ceiling. A maximal
// attachment must reach the turn, not fail the RPC with ResourceExhausted.
func TestGatewayAcceptsAnInboundAttachmentOverTheDefaultGRPCLimit(t *testing.T) {
rec := &recordingGatewayContract{}
ln, err := (&net.ListenConfig{}).Listen(t.Context(), "tcp", "127.0.0.1:0")
if err != nil {
t.Fatalf("listen: %v", err)
}
opts, _, err := gatewayServerOpts(ln.Addr().String(), "")
if err != nil {
t.Fatalf("gatewayServerOpts: %v", err)
}
ctx, cancel := context.WithCancel(t.Context())
defer cancel()
serveCh := make(chan error, 1)
go func() { serveCh <- serveGateway(ctx, ln, rec, nil, opts) }()
defer func() {
cancel()
if err := <-serveCh; err != nil {
t.Errorf("serveGateway: %v", err)
}
}()
conn, err := grpc.NewClient("passthrough:///"+ln.Addr().String(), grpc.WithTransportCredentials(insecure.NewCredentials()))
if err != nil {
t.Fatalf("dial: %v", err)
}
defer conn.Close()
raw := make([]byte, 5<<20) // larger than gRPC's 4 MB default receive limit
stream, err := pb.NewChatServiceClient(conn).Stream(t.Context(), &pb.StreamRequest{
Message: &pb.Message{Media: []*pb.Media{{Type: "image", MimeType: "image/jpeg", Data: raw}}},
})
if err != nil {
t.Fatalf("Stream with a %d-byte attachment: %v", len(raw), err)
}
if _, err := stream.Recv(); !errors.Is(err, io.EOF) {
t.Fatalf("Recv after an empty stream: %v, want EOF", err)
}
rec.mu.Lock()
got := rec.last
rec.mu.Unlock()
if len(got.Media) != 1 {
t.Fatalf("turn received %d attachments, want 1", len(got.Media))
}
if len(got.Media[0].Data) != len(raw) {
t.Fatalf("turn received %d attachment bytes, want %d", len(got.Media[0].Data), len(raw))
}
}
+83 -16
View File
@@ -198,7 +198,7 @@ func npmCacheServerEnv(command, workDir string) []string {
return mcp.NpmCacheEnv(command, filepath.Join(workDir, "mcp-npm-cache"))
}
func configuredMCPProvider(server config.MCPServer, workDir string) (toolprovider.Engine, error) {
func configuredMCPProvider(server config.MCPServer, workDir string, sampling mcp.SamplingHandler) (toolprovider.Engine, error) {
name := strings.TrimSpace(server.Name)
if name == "" {
return nil, fmt.Errorf("MCP server name is required")
@@ -208,6 +208,8 @@ func configuredMCPProvider(server config.MCPServer, workDir string) (toolprovide
transportType = "stdio"
}
samplingOption := mcptoolprovider.WithSamplingHandler(sampling)
switch transportType {
case "stdio":
command := strings.TrimSpace(server.Command)
@@ -220,7 +222,7 @@ func configuredMCPProvider(server config.MCPServer, workDir string) (toolprovide
Dir: server.WorkDir,
Env: npmCacheServerEnv(command, workDir),
})
return mcptoolprovider.New(name, transport, server.ParallelToolCalls), nil
return mcptoolprovider.New(name, transport, server.ParallelToolCalls, samplingOption), nil
case "http", "streamablehttp":
url := strings.TrimSpace(server.URL)
@@ -231,7 +233,7 @@ func configuredMCPProvider(server config.MCPServer, workDir string) (toolprovide
Endpoint: url,
Headers: server.Headers,
})
return mcptoolprovider.New(name, transport, server.ParallelToolCalls), nil
return mcptoolprovider.New(name, transport, server.ParallelToolCalls, samplingOption), nil
case "sse":
sseEndpoint := strings.TrimSpace(server.SSEEndpoint)
@@ -243,7 +245,7 @@ func configuredMCPProvider(server config.MCPServer, workDir string) (toolprovide
MessageEndpoint: strings.TrimSpace(server.MessageEndpoint),
Headers: server.Headers,
})
return mcptoolprovider.New(name, transport, server.ParallelToolCalls), nil
return mcptoolprovider.New(name, transport, server.ParallelToolCalls, samplingOption), nil
default:
return nil, fmt.Errorf("MCP transport %q is not supported", transportType)
@@ -379,6 +381,7 @@ func Run() int { //nolint:cyclop,funlen // the composition root's setup sequence
return 0
}
b.loadCatalog(ctx, args.cfgPath)
b.seedSoul(args.cfgPath)
b.setupObservability(ctx)
if err := b.setupBackends(ctx); err != nil {
@@ -388,6 +391,11 @@ func Run() int { //nolint:cyclop,funlen // the composition root's setup sequence
if err := b.setupLLMAndChat(ctx); err != nil {
return 1
}
// The catalog's consumers exist from here on, and the daemon is the
// process that owns the published configuration, so this is where the
// live catalog refresh belongs. The standalone Gateway reads the catalog
// once at boot and runs no control-plane watches (RunGateway).
b.startModelCatalogRefresh(ctx)
if err := b.buildTreesAndIdentities(ctx); err != nil {
return 1
}
@@ -404,6 +412,9 @@ func Run() int { //nolint:cyclop,funlen // the composition root's setup sequence
if err := b.loadPlugins(); err != nil {
return 1
}
// Every directory loader has run by here, so the reconciliation seeds the
// running set and from its first tick loads only what appears or changes.
b.startPluginReconcile(ctx)
if err := b.registerNATSRPC(); err != nil {
return 1
@@ -414,7 +425,6 @@ func Run() int { //nolint:cyclop,funlen // the composition root's setup sequence
return 1
}
b.setupGuardrails()
if err := b.registerTools(ctx); err != nil {
return 1
}
@@ -560,7 +570,7 @@ func (a chatTaskWriterAdapter) EnqueueChatTask(
type chatTaskControllerAdapter struct {
taskByID func(context.Context, int64) (*workflow.Task, error)
requeue func(context.Context, int64, string, string) error
approve func(context.Context, *string, taskactions.Actor, int64, taskactions.ActionPayload) error
cancelExecution func(context.Context, int64, string, string) ([]int64, error)
}
@@ -578,8 +588,16 @@ func (a chatTaskControllerAdapter) ChatTaskStatus(ctx context.Context, taskID in
return gateway.ChatTaskStatus{}, false, fmt.Errorf("task %d is not chat-originated", taskID)
}
func (a chatTaskControllerAdapter) ApproveChatTask(ctx context.Context, taskID int64) error {
return a.requeue(ctx, taskID, workflow.StatusWaitingHuman, "implement")
func (a chatTaskControllerAdapter) ApproveChatTask(ctx context.Context, taskID int64, actor taskactions.Actor) error {
if a.approve == nil {
return fmt.Errorf("task approval is unavailable")
}
// The chat-bound identity is both the scope the task must belong to and
// the actor the record names: a chat command has no cross-identity
// authority, and /approve carries no selection syntax, so the review gate
// answer posts every offered finding.
scope := string(actor.Identity)
return a.approve(ctx, &scope, actor, taskID, taskactions.ActionPayload{})
}
func (a chatTaskControllerAdapter) CancelChatTask(ctx context.Context, taskID int64, reason string) error {
@@ -684,15 +702,15 @@ type chatTaskActorAdapter struct {
}
func (a chatTaskActorAdapter) ApplyChatTaskAction(
ctx context.Context, identity *string, actor taskactions.Actor, taskID int64, action taskstate.Action,
ctx context.Context, identity *string, actor taskactions.Actor, taskID int64, action taskstate.Action, res taskactions.ActionPayload,
) (gateway.TaskActionResult, error) {
if a.contract == nil {
return gateway.TaskActionResult{}, gateway.ErrChatCapabilityUnavailable
}
if identity == nil {
return a.contract.ApplyOperatorTaskAction(ctx, actor, taskID, action)
return a.contract.ApplyOperatorTaskAction(ctx, actor, taskID, action, res)
}
return a.contract.ApplyTaskAction(ctx, *identity, taskID, action)
return a.contract.ApplyTaskAction(ctx, *identity, taskID, action, res)
}
func chatTaskProfiles(cfg config.Config) ([]gateway.TaskProfile, string) {
@@ -955,10 +973,12 @@ func safePluginInfo(p plugin.Plugin) (name, version string) {
// there is deliberately no host execution fallback.
func startContainers(
ctx context.Context,
cfg config.Config,
cfgHolder *config.Holder,
secrets *secret.Registry,
log *slog.Logger,
) (*container.Pool, storage.Backend, func()) {
noop := func() {}
cfg := cfgHolder.Get()
// Single Docker client shared between pool and storage backend.
dockerCli, err := client.New(client.FromEnv)
if err != nil {
@@ -976,16 +996,15 @@ func startContainers(
MaxConcurrency: cfg.Containers.MaxConcurrency,
MaxUptime: cfg.Containers.MaxUptime.Std(),
PullPolicy: cfg.Containers.PullPolicy,
RegistryAuth: resolveRegistryAuth(cfg.Containers.RegistryAuth, secrets, log),
Network: cfg.Containers.Network,
DockerClient: dockerCli,
// Only the embedded broker binds a discovered host gateway. The
// standalone State Store uses its configured, container-reachable target.
RequireHostGateway: cfg.NATS.Mode == config.NATSModeEmbedded,
}, log)
}, containerLiveSettings(cfgHolder), log)
if err != nil {
// A missing image is recoverable by hand. The daemon sends no registry
// credentials on pull (internal/container/pool.go), so a private
// registry always needs the operator's CLI to fetch it first.
// A missing image is recoverable by hand.
log.Error("autonomous workflows unavailable: container pool unavailable", "err", err,
"hint", "run `docker compose pull agent` (or `build agent`) so the image is present locally")
return nil, storage.NewDockerBackend(dockerCli), closeDocker
@@ -1001,3 +1020,51 @@ func startContainers(
closeDocker()
}
}
// containerLiveSettings reads the per-acquire container settings from the
// running config. The pool calls it on every acquire, so a stored
// container-runtime-policies change reaches the next container without a
// restart. Only fields the resource document carries are taken live:
// RegistryAuth is file-owned and stays the boot-resolved value, DockerClient
// and RequireHostGateway are construction-time, and GracePeriod is not part
// of the document.
func containerLiveSettings(cfgHolder *config.Holder) func() container.Config {
return func() container.Config {
cfg := cfgHolder.Get()
return container.Config{
Image: cfg.Containers.Image,
MaxConcurrency: cfg.Containers.MaxConcurrency,
MaxUptime: cfg.Containers.MaxUptime.Std(),
PullPolicy: cfg.Containers.PullPolicy,
Network: cfg.Containers.Network,
}
}
}
// resolveRegistryAuth resolves [containers].registry_auth into the credential
// the container pool sends on pull. It returns "" -- an anonymous pull, the
// pool's behaviour before the field existed -- whenever the credential is not
// configured or cannot be resolved, because a missing registry credential must
// not stop autonomous work: the pool only needs it for a private registry. The
// warning is what makes the 401 that a private registry will still answer
// explicable after a typo or an unset secret. A ref that names only one half is
// a config error caught by configuration.Validate, not a missing credential.
func resolveRegistryAuth(ref secret.SecretRef, secrets *secret.Registry, log *slog.Logger) string {
if ref == (secret.SecretRef{}) {
return ""
}
if secrets == nil {
log.Warn("registry credential unavailable; pulling without auth", "reason", "no secret registry", "engine", ref.Engine, "key", ref.Key)
return ""
}
value, err := secrets.Resolve(ref)
if err != nil {
log.Warn("registry credential unavailable; pulling without auth", "err", err, "engine", ref.Engine, "key", ref.Key)
return ""
}
if strings.TrimSpace(value) == "" {
log.Warn("registry credential resolved empty; pulling without auth", "engine", ref.Engine, "key", ref.Key)
return ""
}
return value
}
+102 -9
View File
@@ -22,14 +22,17 @@ import (
"github.com/samcharles93/archie-core/internal/config"
"github.com/samcharles93/archie-core/internal/daemon"
"github.com/samcharles93/archie-core/internal/domain/messaging"
"github.com/samcharles93/archie-core/internal/domain/taskactions"
"github.com/samcharles93/archie-core/internal/domain/workflow"
"github.com/samcharles93/archie-core/internal/events"
"github.com/samcharles93/archie-core/internal/forge"
"github.com/samcharles93/archie-core/internal/forgerpc"
"github.com/samcharles93/archie-core/internal/gateway"
"github.com/samcharles93/archie-core/internal/infrastructure/postgres/pgstore"
taskactionstore "github.com/samcharles93/archie-core/internal/infrastructure/taskactions"
"github.com/samcharles93/archie-core/internal/logging"
"github.com/samcharles93/archie-core/internal/secret"
"github.com/samcharles93/archie-core/internal/taskstate"
"github.com/samcharles93/archie-core/internal/taskstate/taskstatetest"
"github.com/samcharles93/archie-core/internal/tools"
"github.com/samcharles93/archie-core/internal/worktree"
@@ -199,6 +202,65 @@ func TestResolveForgeBuildsClientWhenTokenResolves(t *testing.T) {
}
}
// resolveRegistryAuth hands the container pool the resolved
// [containers].registry_auth, and degrades to an anonymous pull (with a
// warning) whenever there is nothing resolvable to hand it. A deployment with
// no private registry must be unaffected by the field existing, and a private
// registry whose secret is missing or misnamed must still let every task that
// does not need it run -- the pool simply pulls without auth, as it always did,
// and the warning explains the 401 that follows.
func TestResolveRegistryAuthDegradesRatherThanFailing(t *testing.T) {
t.Run("unconfigured is silent", func(t *testing.T) {
rec := &recordingHandler{}
if got := resolveRegistryAuth(secret.SecretRef{}, secret.NewRegistry(), slog.New(rec)); got != "" {
t.Errorf("resolveRegistryAuth(empty) = %q, want empty", got)
}
if warns := rec.Warnings(); len(warns) != 0 {
t.Errorf("unconfigured registry auth logged %v, want nothing", warns)
}
})
t.Run("resolved credential reaches the pool", func(t *testing.T) {
t.Setenv("ARCHIE_TEST_REGISTRY_AUTH", `{"username":"bot","password":"s3cret"}`)
rec := &recordingHandler{}
got := resolveRegistryAuth(
secret.SecretRef{Engine: "env", Key: "ARCHIE_TEST_REGISTRY_AUTH"},
secret.NewRegistry(), slog.New(rec))
if want := `{"username":"bot","password":"s3cret"}`; got != want {
t.Errorf("resolveRegistryAuth = %q, want the resolved secret %q", got, want)
}
if warns := rec.Warnings(); len(warns) != 0 {
t.Errorf("a resolvable credential logged %v, want nothing", warns)
}
})
t.Run("unresolvable credential degrades with a warning", func(t *testing.T) {
rec := &recordingHandler{}
got := resolveRegistryAuth(
secret.SecretRef{Engine: "env", Key: "ARCHIE_TEST_REGISTRY_AUTH_DEFINITELY_UNSET"},
secret.NewRegistry(), slog.New(rec))
if got != "" {
t.Errorf("resolveRegistryAuth = %q, want empty for an unresolvable credential", got)
}
warns := rec.Warnings()
if len(warns) != 1 || !strings.Contains(warns[0], "registry credential unavailable") {
t.Errorf("warnings = %v, want exactly one saying the credential was unavailable", warns)
}
})
t.Run("no secret registry degrades instead of panicking", func(t *testing.T) {
rec := &recordingHandler{}
got := resolveRegistryAuth(
secret.SecretRef{Engine: "env", Key: "ANYTHING"}, nil, slog.New(rec))
if got != "" {
t.Errorf("resolveRegistryAuth = %q, want empty with no secret registry", got)
}
if warns := rec.Warnings(); len(warns) != 1 {
t.Errorf("warnings = %v, want exactly one", warns)
}
})
}
func TestResolveProviderSecretSetsPrivateRuntimeEnvironment(t *testing.T) {
registry := secret.NewRegistry()
registry.Register(providerSecretEngine{value: "provider-secret"})
@@ -431,7 +493,7 @@ func TestConfiguredMCPProviderSupportsAllTransportTypes(t *testing.T) {
}
for _, tt := range tests {
t.Run(tt.name, func(t *testing.T) {
provider, err := configuredMCPProvider(tt.server, t.TempDir())
provider, err := configuredMCPProvider(tt.server, t.TempDir(), nil)
if (err != nil) != tt.wantErr {
t.Fatalf("configuredMCPProvider() error = %v, wantErr %t", err, tt.wantErr)
}
@@ -670,11 +732,19 @@ func TestChatTaskControllerAdapterRejectsForgeTask(t *testing.T) {
func TestChatTaskControllerAdapterTransitions(t *testing.T) {
task := &workflow.Task{ID: 42, Source: workflow.SourceChat, Status: workflow.StatusWaitingHuman}
var requeueFrom, requeueWorkflow, cancelReason, cancelTo string
var approveScope string
var approveActor taskactions.Actor
var approveRes taskactions.ActionPayload
var cancelReason, cancelTo string
adapter := chatTaskControllerAdapter{
taskByID: func(context.Context, int64) (*workflow.Task, error) { return task, nil },
requeue: func(_ context.Context, _ int64, from, workflow string) error {
requeueFrom, requeueWorkflow = from, workflow
// Chat's approval reaches the daemon's one task-action service, not
// its own requeue: this records what the adapter hands it.
approve: func(_ context.Context, scope *string, actor taskactions.Actor, _ int64, res taskactions.ActionPayload) error {
if scope != nil {
approveScope = *scope
}
approveActor, approveRes = actor, res
return nil
},
cancelExecution: func(_ context.Context, _ int64, reason, to string) ([]int64, error) {
@@ -683,11 +753,17 @@ func TestChatTaskControllerAdapterTransitions(t *testing.T) {
},
}
if err := adapter.ApproveChatTask(context.Background(), task.ID); err != nil {
if err := adapter.ApproveChatTask(context.Background(), task.ID, taskactions.ActorFromScope("archie")); err != nil {
t.Fatalf("ApproveChatTask(): %v", err)
}
if requeueFrom != workflow.StatusWaitingHuman || requeueWorkflow != "implement" {
t.Errorf("requeue = %q/%q, want waiting_human/implement", requeueFrom, requeueWorkflow)
// The scope is the chat-bound identity, and /approve carries no
// instruction or selection syntax, so the review-gate payload is empty
// (an absent selection means every offered finding).
if approveScope != "archie" || approveActor.Identity != "archie" {
t.Errorf("approve scope/actor = %q/%q, want the chat-bound identity", approveScope, approveActor.Identity)
}
if approveRes.Instructions != "" || len(approveRes.Findings) != 0 {
t.Errorf("approve payload = %+v, want the zero review response for chat", approveRes)
}
if err := adapter.CancelChatTask(context.Background(), task.ID, "cancelled by test"); err != nil {
t.Fatalf("CancelChatTask(): %v", err)
@@ -834,9 +910,15 @@ func TestChatTaskCommandsEndToEnd(t *testing.T) {
chatTaskWriterAdapter{enqueue: st.EnqueueChatTask},
profiles,
)
// Chat's /approve reaches the same task-action service the daemon wires
// (docs/prds/pr-review-operator-response.md, Decision 1), over the real
// store this test uses.
actions := taskactionstore.NewService(taskactionstore.Store{TaskStore: st}, nil, nil, nil, nil, nil, nil, nil)
controller := gateway.NewStoreTaskController(chatTaskControllerAdapter{
taskByID: st.TaskByID,
requeue: st.Requeue,
taskByID: st.TaskByID,
approve: func(ctx context.Context, scope *string, actor taskactions.Actor, taskID int64, res taskactions.ActionPayload) error {
return actions.Apply(ctx, scope, actor, taskID, taskstate.ActionApprove, res)
},
cancelExecution: st.CancelExecution,
})
router := gateway.NewRouter(st, nil, "test")
@@ -862,6 +944,17 @@ func TestChatTaskCommandsEndToEnd(t *testing.T) {
t.Fatalf("spawned task = %+v", task)
}
// Feasibility's deliver stage names implement on the task before it waits
// (docs/prds/pr-review-operator-response.md, Decision 1), so the wait the
// seed models carries that name; the approval resumes it.
waiting, err := st.TaskByID(ctx, taskID)
if err != nil || waiting == nil {
t.Fatalf("task before wait = (%+v, %v)", waiting, err)
}
waiting.Workflow = "implement"
if err := st.Update(ctx, waiting); err != nil {
t.Fatalf("name the resuming workflow: %v", err)
}
taskstatetest.Seed(t.Context(), t, st, taskID, workflow.StatusQueued, workflow.StatusWaitingHuman, "await approval")
task, err = st.TaskByID(ctx, taskID)
if err != nil || task == nil || task.Status != workflow.StatusWaitingHuman {
+192
View File
@@ -0,0 +1,192 @@
package archied
import (
"context"
"errors"
"fmt"
"reflect"
"slices"
"strings"
"github.com/samcharles93/archie-core/internal/config"
"github.com/samcharles93/archie-core/internal/tools/minimax"
toolprovider "github.com/samcharles93/archie-core/internal/tools/provider"
"github.com/samcharles93/archie-core/internal/tools/webfetch"
)
// appliedMCPServer is one MCP server the running provider registry was built
// from: the configuration it was built with and the manifest id it registered
// under. A live tool-settings change diffs the stored server set against this
// record, so a server that did not change is left running and untouched
// (archie-core-zfb0.3).
type appliedMCPServer struct {
server config.MCPServer
id string
}
// buildMCPProvider builds the tool-provider engine for one configured MCP
// server. boot.mcpProvider overrides it in tests; production uses
// configuredMCPProvider with the running config's work directory and the
// daemon's sampling handler. Boot registration and the live reconciliation
// both reach the engine through here, so the two cannot build one differently.
func (b *boot) buildMCPProvider(srv config.MCPServer) (toolprovider.Engine, error) {
if b.mcpProvider != nil {
return b.mcpProvider(srv)
}
return configuredMCPProvider(srv, b.cfgHolder.Get().WorkDir, b.mcpSamplingHandler())
}
// reconcileToolSettings applies a live tool-settings change to the components
// built once from that kind: the MCP provider set (diffed, so an unchanged
// server keeps running), the web_fetch and minimax tool entries (rebuilt,
// because the entry captured the config at construction) and the tool-output
// policy (whose spill directory is ensured here; the task snapshot reads the
// policy fresh per dispatch and boot.toolLimits is read per call). A nil tool
// registry means this process built no tools, so only the config was layered.
func (b *boot) reconcileToolSettings(ctx context.Context, cfg config.Config) error {
if err := toolLimits(cfg).EnsureSpillDir(); err != nil {
b.log.Warn("tool spill directory unavailable; large results will be truncated instead", "err", err)
}
var problems []error
if err := b.reconcileMCPServers(ctx, cfg.Tools.MCPServers); err != nil {
problems = append(problems, err)
}
if b.toolReg != nil {
// Unregister first: the boot entry is being replaced, and a second
// registration would collide on the tool name. Minimax resolves its
// API key again as part of the rebuild, so a rotated credential
// applies too.
b.toolReg.Unregister(webfetch.ToolName, minimax.ToolName)
b.registerWebFetchTool(cfg)
b.registerMinimaxTool(cfg, b.log)
}
return errors.Join(problems...)
}
// reconcileMCPServers brings the running provider set in line with the MCP
// servers the layered configuration now names: a server that appeared
// connects, one that disappeared disconnects, one whose configuration changed
// reconnects, and an unchanged one is left running and untouched. It is
// idempotent, so it can run on every stored tool-settings version without
// disturbing the servers that did not change.
//
// A server whose engine cannot be built or started is reported and left out;
// the servers already running keep running, the degrade-and-skip rule the boot
// registration applies. A changed server whose new engine fails to start is
// rolled back to the old one by the registry, so a refused change does not
// leave the server down.
func (b *boot) reconcileMCPServers(ctx context.Context, servers []config.MCPServer) error {
if b.providerRegistry == nil {
return nil
}
b.mcpMu.Lock()
defer b.mcpMu.Unlock()
desired := make(map[string]config.MCPServer, len(servers))
for _, srv := range servers {
if name := strings.TrimSpace(srv.Name); name != "" {
desired[name] = srv
}
}
state := b.mcpRegistryState()
var problems []error
problems = append(problems, b.disconnectAbsentMCPServers(ctx, desired)...)
problems = append(problems, b.applyDesiredMCPServers(ctx, desired, state)...)
return errors.Join(problems...)
}
// mcpRegistryState is the read side of the provider registry a diff needs: is
// the family running at all, which ids are registered, and which are live.
type mcpRegistryState struct {
familyRunning bool
registered map[string]bool
live map[string]bool
}
func (b *boot) mcpRegistryState() mcpRegistryState {
registered := make(map[string]bool)
for _, id := range b.providerRegistry.RegisteredIDs() {
registered[id] = true
}
live := make(map[string]bool)
for _, id := range b.providerRegistry.RunningIDs() {
live[id] = true
}
return mcpRegistryState{
familyRunning: b.providerRegistry.Running(),
registered: registered,
live: live,
}
}
// disconnectAbsentMCPServers removes every applied server the stored document
// no longer names.
func (b *boot) disconnectAbsentMCPServers(ctx context.Context, desired map[string]config.MCPServer) []error {
var problems []error
for name := range b.mcpApplied {
if _, ok := desired[name]; ok {
continue
}
applied := b.mcpApplied[name]
if err := b.providerRegistry.Remove(ctx, applied.id); err != nil {
problems = append(problems, fmt.Errorf("disconnect MCP server %q: %w", name, err))
continue
}
delete(b.mcpApplied, name)
}
return problems
}
// applyDesiredMCPServers connects, reconnects or restarts the servers the
// stored document names, skipping the ones already running unchanged.
func (b *boot) applyDesiredMCPServers(ctx context.Context, desired map[string]config.MCPServer, state mcpRegistryState) []error {
names := make([]string, 0, len(desired))
for name := range desired {
names = append(names, name)
}
slices.Sort(names)
var problems []error
for _, name := range names {
srv := desired[name]
if b.mcpServerCurrent(name, srv, state) {
continue
}
engine, err := b.buildMCPProvider(srv)
if err != nil {
problems = append(problems, fmt.Errorf("MCP server %q: %w", name, err))
continue
}
if err := b.applyMCPProvider(ctx, engine, state); err != nil {
problems = append(problems, fmt.Errorf("connect MCP server %q: %w", name, err))
continue
}
b.mcpApplied[name] = appliedMCPServer{server: srv, id: engine.Manifest().ID}
}
return problems
}
// mcpServerCurrent reports whether the server is already running unchanged, so
// a reconciliation can leave it untouched. Present means running while the
// family runs, or still registered while it has not started; a
// registered-but-not-running server on a running family is a refused earlier
// start and is worth another attempt.
func (b *boot) mcpServerCurrent(name string, srv config.MCPServer, state mcpRegistryState) bool {
applied, ok := b.mcpApplied[name]
if !ok || !reflect.DeepEqual(applied.server, srv) {
return false
}
if !state.registered[applied.id] {
return false
}
return !state.familyRunning || state.live[applied.id]
}
// applyMCPProvider replaces a registered provider and adds an absent one,
// whichever the registry needs for that id.
func (b *boot) applyMCPProvider(ctx context.Context, engine toolprovider.Engine, state mcpRegistryState) error {
if state.registered[engine.Manifest().ID] {
return b.providerRegistry.Replace(ctx, engine)
}
return b.providerRegistry.Add(ctx, engine)
}
+71
View File
@@ -0,0 +1,71 @@
package archied
import (
"context"
"errors"
"fmt"
"github.com/samcharles93/ai-sdk/core"
"github.com/samcharles93/archie-core/internal/agentexec"
"github.com/samcharles93/archie-core/internal/tools/mcp"
)
// mcpSamplingHandler answers an MCP server's sampling/createMessage request
// from the daemon's own chat model, so a server can ask for a completion
// without archie opening a second provider path. A missing runtime or model
// is a refused request, never a dropped one: the client maps the returned
// error to a JSON-RPC error response for that request alone, leaving the
// server session intact.
func (b *boot) mcpSamplingHandler() mcp.SamplingHandler {
return func(ctx context.Context, req mcp.SamplingRequest) (mcp.SamplingResult, error) {
llm := b.chatLLM()
if llm == nil {
return mcp.SamplingResult{}, errors.New("mcp sampling: no model runtime is configured")
}
if b.chatModels == nil {
return mcp.SamplingResult{}, errors.New("mcp sampling: no chat model is configured")
}
model := b.chatModels.ActiveModel()
if model == "" {
return mcp.SamplingResult{}, errors.New("mcp sampling: no chat model is configured")
}
messages, err := agentexec.SamplingMessages(req)
if err != nil {
return mcp.SamplingResult{}, err
}
options := core.GenerateOptions{
Messages: messages,
System: req.SystemPrompt,
MaxSteps: 1,
MaxTokens: samplingMaxTokens(req, b.chatModels, model),
}
if req.Temperature != nil {
options.Temperature = float32(*req.Temperature)
}
result, err := llm.Chat(ctx, model, options)
// A sampling call is a model call this process made, so it is
// recorded for /status like every other one; a nil recorder records
// nothing.
b.providerOutcomes.record(model, err)
if err != nil {
return mcp.SamplingResult{}, fmt.Errorf("mcp sampling: %w", err)
}
return mcp.SamplingResult{
Model: model,
Role: "assistant",
Content: mcp.SamplingContent{Type: "text", Text: result.Text},
StopReason: string(result.FinishReason),
}, nil
}
}
// samplingMaxTokens bounds the completion to the request's own maxTokens,
// falling back to a conservative default, and never past the model's own
// output ceiling. The bound itself is shared with the agent worker's sampling
// path (agentexec.SamplingMaxTokens) so the same server sees the same bound on
// either path.
func samplingMaxTokens(req mcp.SamplingRequest, models *chatModelManager, model string) int {
details, ok := models.ModelDetails(model)
return agentexec.SamplingMaxTokens(req.MaxTokens, details.MaxOutputTokens, ok && details.Reasoning)
}
+98
View File
@@ -0,0 +1,98 @@
package archied
import (
"io"
"net/http"
"net/http/httptest"
"strings"
"testing"
"github.com/samcharles93/archie-core/internal/agentexec"
"github.com/samcharles93/archie-core/internal/tools/mcp"
)
func newSamplingTestRuntime(t *testing.T) *httptest.Server {
t.Helper()
api := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {
w.Header().Set("Content-Type", "application/json")
_, _ = io.WriteString(w, `{"id":"chatcmpl-1","object":"chat.completion","created":1,"model":"gpt-5.6",`+
`"choices":[{"index":0,"message":{"role":"assistant","content":"four"},"finish_reason":"stop"}],`+
`"usage":{"prompt_tokens":1,"completion_tokens":1,"total_tokens":2}}`)
}))
t.Cleanup(api.Close)
return api
}
// TestMCPSamplingHandlerAnswersFromConfiguredModel proves a server-initiated
// sampling request is answered by the daemon's own chat model and recorded on
// /status like every other model call this process makes.
func TestMCPSamplingHandlerAnswersFromConfiguredModel(t *testing.T) {
api := newSamplingTestRuntime(t)
b := newBootstrap()
b.setLLM(agentexec.NewRuntime(map[string]agentexec.Provider{
"openai": {Class: "openai", BaseURL: api.URL},
}))
b.chatModels = newChatModelManager(map[string]string{"chat": "openai/gpt-5.6"})
b.providerOutcomes = newProviderOutcomeRecorder()
result, err := b.mcpSamplingHandler()(t.Context(), mcp.SamplingRequest{
Messages: []mcp.SamplingMessage{{
Role: "user",
Content: mcp.SamplingContent{Type: "text", Text: "2+2?"},
}},
MaxTokens: 16,
})
if err != nil {
t.Fatalf("sampling handler = %v, want a completion", err)
}
if result.Model != "openai/gpt-5.6" {
t.Errorf("result.Model = %q, want the configured chat model", result.Model)
}
if result.Role != "assistant" || result.Content.Type != "text" || result.Content.Text != "four" {
t.Errorf("result = %+v, want an assistant text completion", result)
}
outcome, attempted := b.providerOutcomes.LastChatModelOutcome()
if !attempted || outcome.Model != "openai/gpt-5.6" || outcome.Err != "" {
t.Errorf("recorded outcome = %+v (attempted %v), want the sampling call recorded as successful", outcome, attempted)
}
}
// TestMCPSamplingHandlerRefusesCleanlyWithNoModel proves a daemon with no
// usable model answers the request with an error instead of panicking or
// dropping it: the server sees a failed request, not a dead session.
func TestMCPSamplingHandlerRefusesCleanlyWithNoModel(t *testing.T) {
tests := []struct {
name string
wired bool
want string
}{
{name: "no model runtime", wired: false, want: "no model runtime"},
{name: "no chat model configured", wired: true, want: "no chat model"},
}
for _, tc := range tests {
t.Run(tc.name, func(t *testing.T) {
b := newBootstrap()
if tc.wired {
b.setLLM(agentexec.NewRuntime(map[string]agentexec.Provider{
"openai": {Class: "openai"},
}))
}
b.chatModels = newChatModelManager(nil)
_, err := b.mcpSamplingHandler()(t.Context(), mcp.SamplingRequest{
Messages: []mcp.SamplingMessage{{
Role: "user",
Content: mcp.SamplingContent{Type: "text", Text: "hi"},
}},
})
if err == nil {
t.Fatal("sampling handler returned no error with no usable model")
}
if !strings.Contains(err.Error(), tc.want) {
t.Errorf("error = %q, want it to mention %q", err, tc.want)
}
})
}
}
+1 -1
View File
@@ -22,7 +22,7 @@ func applyModelCatalog(cfg *config.Config, snapshot modelcatalog.Snapshot) []str
for _, model := range provider.Models {
ref := provider.ID + "/" + model.ID
models = append(models, ref)
limits[ref] = config.ModelLimits{ContextWindow: model.ContextWindow, MaxOutputTokens: model.MaxOutputTokens}
limits[ref] = config.ModelLimits{ContextWindow: model.ContextWindow, MaxOutputTokens: model.MaxOutputTokens, Reasoning: model.Reasoning}
}
}
cfg.ModelLimits = limits
@@ -0,0 +1,250 @@
package archied
import (
"context"
"encoding/json"
"errors"
"io/fs"
"net/http"
"net/http/httptest"
"os"
"path/filepath"
"slices"
"strings"
"sync/atomic"
"testing"
"time"
"github.com/samcharles93/archie-core/internal/agentexec"
"github.com/samcharles93/archie-core/internal/infrastructure/modelcatalog"
"github.com/samcharles93/archie-core/internal/secret"
"github.com/samcharles93/archie-core/internal/webui"
)
// catalogServer answers the model catalog read with a models.dev-shaped
// document. It is the outside dependency the refresh loop actually meets:
// modelcatalog.Load fetches this and caches what it read.
func catalogServer(t *testing.T, document func() map[string]any) *httptest.Server {
t.Helper()
srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {
w.Header().Set("Content-Type", "application/json")
_ = json.NewEncoder(w).Encode(document())
}))
t.Cleanup(srv.Close)
return srv
}
// catalogProvider is one usable provider entry: an environment variable that
// resolves (so modelcatalog keeps it) and the tool-callable models it offers.
func catalogProvider(id, env string, models ...string) map[string]any {
entries := make(map[string]any, len(models))
for _, model := range models {
entries[model] = map[string]any{
"id": model, "name": strings.ToUpper(model), "tool_call": true,
"limit": map[string]any{"context": 128000, "output": 16000},
}
}
return map[string]any{
"id": id, "name": strings.ToUpper(id), "npm": "@ai-sdk/openai-compatible",
"api": "https://" + id + ".test/v1", "env": []string{env}, "models": entries,
}
}
// newCatalogBoot builds the smallest boot the refresh path runs against: the
// control plane reads behind runtimeConfig, the secrets registry the catalog
// resolves keys through, the chat runtime a refresh must reach, and a catalog
// endpoint a test controls.
func newCatalogBoot(t *testing.T, url string) *boot {
t.Helper()
b := newReloadBoot(t, &controlPlaneStub{values: databaseOwnedResources()})
b.secrets = secret.NewRegistry()
b.chatModels = newChatModelManager(fileConfig().Models)
b.setLLM(agentexec.NewRuntime(executionProviders(fileConfig())))
b.catalogURL = url
b.catalogCachePath = filepath.Join(t.TempDir(), "models.json")
if err := b.loadRuntimeConfig(t.Context()); err != nil {
t.Fatalf("loadRuntimeConfig: %v", err)
}
return b
}
func awaitCatalog(t *testing.T, b *boot, ref string) {
t.Helper()
deadline := time.Now().Add(5 * time.Second)
for !slices.Contains(b.chatModels.Models(), ref) {
if time.Now().After(deadline) {
t.Fatalf("chat models = %v, want %s to appear within the refresh interval", b.chatModels.Models(), ref)
}
time.Sleep(2 * time.Millisecond)
}
}
// TestRefreshModelCatalogRepublishesEveryProjectionSite: a catalog read after
// boot must reach everything derived from it -- the running config a workflow
// stage resolves model limits from, the dashboard's published catalog view,
// and the chat runtime's model list and details. A value that reaches the
// boot struct alone parses and does nothing.
func TestRefreshModelCatalogRepublishesEveryProjectionSite(t *testing.T) {
srv := catalogServer(t, func() map[string]any {
return map[string]any{"acme": catalogProvider("acme", "ACME_API_KEY", "acme-tool")}
})
t.Setenv("ACME_API_KEY", "resolved-by-the-secret-registry")
b := newCatalogBoot(t, srv.URL)
if snapshot, _ := b.catalogState(); len(snapshot.Providers) != 0 {
t.Fatalf("boot catalog = %+v, want the empty snapshot this test starts from", snapshot.Providers)
}
if err := b.refreshModelCatalog(t.Context()); err != nil {
t.Fatalf("refreshModelCatalog: %v", err)
}
if got := b.cfgHolder.Get().ModelLimits["acme/acme-tool"]; got.ContextWindow != 128000 || got.MaxOutputTokens != 16000 {
t.Errorf("running model limits = %+v, want the refreshed catalog's capacity", got)
}
if _, models := b.catalogState(); !slices.Contains(models, "acme/acme-tool") {
t.Errorf("catalog models = %v, want the refreshed reference", models)
}
view := webui.BuildConfigView(b.configViewInput(t.Context()))
if len(view.Catalog) != 1 || view.Catalog[0].ID != "acme" {
t.Fatalf("published catalog = %+v, want the refreshed provider", view.Catalog)
}
if !slices.Contains(view.Catalog[0].Models, "acme-tool") {
t.Errorf("published catalog models = %v, want the refreshed model", view.Catalog[0].Models)
}
if !slices.Contains(b.chatModels.Models(), "acme/acme-tool") {
t.Errorf("chat models = %v, want the refreshed reference offered", b.chatModels.Models())
}
if _, ok := b.chatModels.ModelDetails("acme/acme-tool"); !ok {
t.Error("chat model details missing the refreshed model")
}
if got := b.chatModels.ProviderDisplayName("acme"); got != "ACME" {
t.Errorf("provider display name = %q, want the refreshed catalog's name", got)
}
}
// TestModelCatalogRefreshLoopPicksUpANewModelWhileTheDaemonRuns is the
// without-a-restart proof: the process is already running on the catalog it
// booted with, and a model that appears upstream afterwards is offered by the
// next tick.
func TestModelCatalogRefreshLoopPicksUpANewModelWhileTheDaemonRuns(t *testing.T) {
var calls atomic.Int64
srv := catalogServer(t, func() map[string]any {
if calls.Add(1) == 1 {
return map[string]any{"acme": catalogProvider("acme", "ACME_API_KEY", "acme-one")}
}
return map[string]any{"acme": catalogProvider("acme", "ACME_API_KEY", "acme-one", "acme-two")}
})
t.Setenv("ACME_API_KEY", "resolved-by-the-secret-registry")
b := newCatalogBoot(t, srv.URL)
if err := b.refreshModelCatalog(t.Context()); err != nil {
t.Fatalf("initial refresh: %v", err)
}
if slices.Contains(b.chatModels.Models(), "acme/acme-two") {
t.Fatal("the second model was offered before upstream published it")
}
b.catalogRefreshInterval = 5 * time.Millisecond
ctx, cancel := context.WithCancel(t.Context())
defer cancel()
b.startModelCatalogRefresh(ctx)
awaitCatalog(t, b, "acme/acme-two")
if got := b.cfgHolder.Get().ModelLimits["acme/acme-two"]; got.ContextWindow != 128000 {
t.Errorf("running model limits for the new model = %+v, want the refreshed capacity", got)
}
}
// TestRefreshModelCatalogKeepsTheLoadedCatalogWhenTheReadFails: the catalog
// service having a bad minute must not blank the model list the process is
// already running on. The failed read is reported; the snapshot stays.
func TestRefreshModelCatalogKeepsTheLoadedCatalogWhenTheReadFails(t *testing.T) {
srv := catalogServer(t, func() map[string]any {
return map[string]any{"acme": catalogProvider("acme", "ACME_API_KEY", "acme-one")}
})
t.Setenv("ACME_API_KEY", "resolved-by-the-secret-registry")
b := newCatalogBoot(t, srv.URL)
if err := b.refreshModelCatalog(t.Context()); err != nil {
t.Fatalf("refreshModelCatalog: %v", err)
}
// The endpoint stops answering and the cache is gone, so the read has no
// fallback left.
srv.Close()
if err := os.Remove(b.catalogCachePath); err != nil && !errors.Is(err, fs.ErrNotExist) {
t.Fatalf("remove cache: %v", err)
}
if err := b.refreshModelCatalog(t.Context()); err == nil {
t.Fatal("refreshModelCatalog error = nil, want the failed read reported")
}
if _, models := b.catalogState(); !slices.Contains(models, "acme/acme-one") {
t.Errorf("catalog models = %v, want the loaded catalog kept", models)
}
if got := b.cfgHolder.Get().ModelLimits["acme/acme-one"]; got.ContextWindow != 128000 {
t.Errorf("running model limits = %+v, want the last-known-good catalog", got)
}
}
// TestBootCatalogReachesTheRunningConfig is the projection-site check for the
// boot half: the catalog boot read must be in the config the process runs on,
// not only in the boot struct the chat runtime was built from. Otherwise the
// first live update is the first snapshot to carry a catalog model limit.
func TestBootCatalogReachesTheRunningConfig(t *testing.T) {
srv := catalogServer(t, func() map[string]any {
return map[string]any{"acme": catalogProvider("acme", "ACME_API_KEY", "acme-one")}
})
t.Setenv("ACME_API_KEY", "resolved-by-the-secret-registry")
b := newCatalogBoot(t, srv.URL)
b.loadCatalog(t.Context(), filepath.Join(t.TempDir(), "config.toml"))
if got := b.cfgHolder.Get().ModelLimits["acme/acme-one"]; got.ContextWindow != 128000 {
t.Errorf("running model limits = %+v, want the catalog boot loaded", got)
}
if _, models := b.catalogState(); !slices.Contains(models, "acme/acme-one") {
t.Errorf("catalog models = %v, want the boot catalog", models)
}
}
// TestChatModelManagerReplacesItsCatalogRatherThanMerging: a refresh must
// drop what the previous snapshot contributed. A manager that only merged
// would keep offering a model the catalog no longer publishes, and would
// describe a provider that no longer exists.
func TestChatModelManagerReplacesItsCatalogRatherThanMerging(t *testing.T) {
manager := newChatModelManager(map[string]string{"chat": "acme/one"})
manager.SetModelCatalog(
modelcatalog.Snapshot{Providers: []modelcatalog.Provider{
{ID: "acme", Name: "Acme", Models: []modelcatalog.Model{{ID: "one"}, {ID: "two"}}},
}},
[]string{"acme/one", "acme/two"},
)
if !slices.Contains(manager.Models(), "acme/two") {
t.Fatalf("chat models = %v, want the catalog's models offered", manager.Models())
}
if got := manager.ProviderDisplayName("acme"); got != "Acme" {
t.Fatalf("provider display name = %q, want the catalog's name", got)
}
manager.SetModelCatalog(
modelcatalog.Snapshot{Providers: []modelcatalog.Provider{
{ID: "acme", Name: "Acme", Models: []modelcatalog.Model{{ID: "one"}}},
}},
[]string{"acme/one"},
)
if slices.Contains(manager.Models(), "acme/two") {
t.Errorf("chat models = %v, want the withdrawn model dropped", manager.Models())
}
if _, ok := manager.ModelDetails("acme/two"); ok {
t.Error("chat model details still carry the withdrawn model")
}
if got := manager.ProviderDisplayName("gone"); got != "" {
t.Errorf("provider display name for an absent provider = %q, want empty", got)
}
if !slices.Contains(manager.Models(), "acme/one") {
t.Errorf("chat models = %v, want the surviving model kept", manager.Models())
}
}
@@ -0,0 +1,163 @@
package archied
import (
"encoding/json"
"io"
"net/http"
"net/http/httptest"
"sync"
"testing"
"github.com/samcharles93/ai-sdk/runtime"
"github.com/samcharles93/archie-core/internal/agentexec"
"github.com/samcharles93/archie-core/internal/gateway"
"github.com/samcharles93/archie-core/internal/infrastructure/modelcatalog"
"github.com/samcharles93/archie-core/internal/tools"
"github.com/samcharles93/archie-core/internal/tools/mcp"
)
// capturedChatRequest is the JSON body the OpenAI-compatible provider sent for
// the last call, which is where the max-token parameter name is chosen.
type capturedChatRequest struct {
mu sync.Mutex
body map[string]any
}
func (c *capturedChatRequest) get() map[string]any {
c.mu.Lock()
defer c.mu.Unlock()
return c.body
}
// newCapturingChatAPI answers every completion and records the wire body, so a
// test can assert the parameter the provider actually emitted rather than an
// option the composition root merely set.
func newCapturingChatAPI(t *testing.T, capture *capturedChatRequest) *httptest.Server {
t.Helper()
api := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
var body map[string]any
_ = json.NewDecoder(r.Body).Decode(&body)
capture.mu.Lock()
capture.body = body
capture.mu.Unlock()
w.Header().Set("Content-Type", "application/json")
_, _ = io.WriteString(w, `{"id":"chatcmpl-1","object":"chat.completion","created":1,"model":"test",`+
`"choices":[{"index":0,"message":{"role":"assistant","content":"ok"},"finish_reason":"stop"}],`+
`"usage":{"prompt_tokens":1,"completion_tokens":1,"total_tokens":2}}`)
}))
t.Cleanup(api.Close)
return api
}
// assertWireBound asserts the max-token parameter the provider emitted: absent
// entirely for a reasoning model (its only accepted form with this ai-sdk
// release), present with the caller's bound for a classic one.
func assertWireBound(t *testing.T, capture *capturedChatRequest, wantBound bool, wantValue int) {
t.Helper()
bound, present := capture.get()["max_tokens"]
if present != wantBound {
t.Fatalf("request body max_tokens present = %v (value %v), want %v; body = %v", present, bound, wantBound, capture.get())
}
if present && bound != float64(wantValue) {
t.Fatalf("request body max_tokens = %v, want %d; body = %v", bound, wantValue, capture.get())
}
}
// TestChatTurnParameterNameFollowsModelClass proves the chat turn path reaches
// the provider without a max-token bound for a reasoning-class model, while a
// classic model still carries one. The bound itself is unchanged; only the
// parameter the chat-completions provider would serialise it into differs, and
// reasoning models reject `max_tokens` with HTTP 400.
func TestChatTurnParameterNameFollowsModelClass(t *testing.T) {
tests := []struct {
name string
reasoning bool
wantBound bool
modelRef string
maxOutput int
wantValue int
}{
{name: "reasoning model", reasoning: true, wantBound: false, modelRef: "openai/gpt-5.4", maxOutput: 4096},
{name: "classic model", reasoning: false, wantBound: true, modelRef: "openai/gpt-4o", maxOutput: 4096, wantValue: 4096},
}
for _, tc := range tests {
t.Run(tc.name, func(t *testing.T) {
capture := &capturedChatRequest{}
api := newCapturingChatAPI(t, capture)
llm := agentexec.NewRuntime(map[string]agentexec.Provider{
"openai": {Class: "openai", BaseURL: api.URL},
})
model := newChatTurnModel(func() *runtime.Runtime { return llm }, tools.NewRegistry(), 1, func() agentexec.ToolLimits { return agentexec.ToolLimits{} }, nil)
prepared, err := model.Prepare(t.Context(), gateway.TurnPrepareContext{
Model: tc.modelRef, Reasoning: tc.reasoning,
})
if err != nil {
t.Fatalf("Prepare() error = %v", err)
}
_, err = prepared.Generate(t.Context(), gateway.TurnModelRequest{
Messages: []gateway.CompressedMessage{{Role: "user", Content: "hi"}},
MaxOutputTokens: tc.maxOutput,
}, nil)
if err != nil {
t.Fatalf("Generate() error = %v", err)
}
assertWireBound(t, capture, tc.wantBound, tc.wantValue)
})
}
}
// TestMCPSamplingParameterNameFollowsModelClass proves the second place the
// process sets a max-token bound -- server-requested sampling, which also
// applies a default when the request is silent -- preserves it for a classic
// model and drops it entirely for a reasoning one.
func TestMCPSamplingParameterNameFollowsModelClass(t *testing.T) {
tests := []struct {
name string
modelRef string
model modelcatalog.Model
wantBound bool
wantValue int
}{
{
name: "reasoning model",
modelRef: "openai/gpt-5.4",
model: modelcatalog.Model{ID: "gpt-5.4", Reasoning: true, MaxOutputTokens: 128000},
},
{
name: "classic model",
modelRef: "openai/gpt-4o",
model: modelcatalog.Model{ID: "gpt-4o", MaxOutputTokens: 128000},
wantBound: true,
wantValue: agentexec.DefaultSamplingMaxTokens,
},
}
for _, tc := range tests {
t.Run(tc.name, func(t *testing.T) {
capture := &capturedChatRequest{}
api := newCapturingChatAPI(t, capture)
b := newBootstrap()
b.setLLM(agentexec.NewRuntime(map[string]agentexec.Provider{
"openai": {Class: "openai", BaseURL: api.URL},
}))
b.chatModels = newChatModelManager(map[string]string{"chat": tc.modelRef})
b.chatModels.SetModelCatalog(modelcatalog.Snapshot{
Providers: []modelcatalog.Provider{{ID: "openai", Models: []modelcatalog.Model{tc.model}}},
}, []string{tc.modelRef})
b.providerOutcomes = newProviderOutcomeRecorder()
_, err := b.mcpSamplingHandler()(t.Context(), mcp.SamplingRequest{
Messages: []mcp.SamplingMessage{{Role: "user", Content: mcp.SamplingContent{Type: "text", Text: "hi"}}},
})
if err != nil {
t.Fatalf("sampling handler error = %v", err)
}
assertWireBound(t, capture, tc.wantBound, tc.wantValue)
})
}
}
+306
View File
@@ -0,0 +1,306 @@
package archied
import (
"context"
"crypto/sha256"
"encoding/hex"
"errors"
"fmt"
"io/fs"
"log/slog"
"os"
"path/filepath"
"slices"
"strings"
"sync"
"time"
"github.com/samcharles93/archie-core/internal/config"
"github.com/samcharles93/archie-core/internal/domain/eda/module"
"github.com/samcharles93/archie-core/internal/plugin"
"github.com/samcharles93/archie-core/internal/plugin/pluginextract"
"github.com/samcharles93/archie-core/internal/secret"
)
// The three directory families a plugin-settings document names and a
// reconciliation reads.
const (
reconcilePlugins = "plugin"
reconcileModules = "module"
reconcileSecrets = "secret"
)
// pluginReconciler loads plugin, module and secret-engine files that appear or
// change in the directories the running configuration names, so an operator can
// drop a file beside a running daemon and have it take effect without a
// restart. Yaegi cannot unload an interpreter, so a removed file's code keeps
// running and the reconciler reports it through apply status until the process
// restarts (docs/prds/plugin-settings-live.md).
//
// Files are content-hashed, so a tick loads only what appeared or changed: the
// boot load's files are seeded as already loaded. A file whose load fails is
// forgotten until its content changes, the same degrade-and-skip rule the boot
// loaders apply.
type pluginReconciler struct {
log *slog.Logger
pluginReconcileTargets
// reconcileMu serializes reconciliations: the poller's tick and a stored
// plugin-settings update both run one, and they share the loaded set.
reconcileMu sync.Mutex
loaded map[string]loadedPluginFile
}
type pluginReconcileTargets struct {
host *plugin.Host
secrets *secret.Registry
modules *module.ModuleRegistry
// dirs returns the running configuration, the source of the three
// directories. It is read each tick so a changed directory retargets the
// next reconciliation without a restart.
dirs func() config.Config
// relayer re-runs the configuration layering after a secret engine loads
// or changes, which re-resolves provider credentials against the new
// engine. Nil in tests that do not layer.
relayer func(ctx context.Context) error
// report publishes the reconciliation outcome, including a nil error on a
// clean pass so an outstanding removal clears.
report func(ctx context.Context, err error)
}
type loadedPluginFile struct {
hash string
category string
// failed marks a file whose last load failed. It is not reported as a
// removal while it is still present, and a content change retries it.
failed bool
}
func newPluginReconciler(log *slog.Logger, targets pluginReconcileTargets) *pluginReconciler {
if log == nil {
log = slog.New(slog.DiscardHandler)
}
return &pluginReconciler{
log: log,
pluginReconcileTargets: targets,
loaded: make(map[string]loadedPluginFile),
}
}
// seed records the files the boot load already evaluated, so the first tick
// loads only what appeared or changed since boot. The daemon calls it after
// loadPlugins, loadWorkflows and configuredSecretRegistry have run.
func (r *pluginReconciler) seed() {
next := make(map[string]loadedPluginFile)
for _, spec := range r.categories(r.dirs()) {
paths, err := goFiles(spec.dir)
if err != nil {
r.log.Warn("plugin directory not seeded", "category", spec.category, "dir", spec.dir, "err", err)
continue
}
for _, path := range paths {
hash, err := fileHash(path)
if err != nil {
continue
}
next[path] = loadedPluginFile{hash: hash, category: spec.category}
}
}
r.reconcileMu.Lock()
r.loaded = next
r.reconcileMu.Unlock()
}
// run reconciles on interval until ctx ends. The interval is the apply-status
// restamp constant, so a dropped file needs no filesystem watch and a
// reconciliation report is refreshed at the same cadence as every other apply
// status record.
func (r *pluginReconciler) run(ctx context.Context, interval time.Duration) {
ticker := time.NewTicker(interval)
defer ticker.Stop()
for {
select {
case <-ctx.Done():
return
case <-ticker.C:
_ = r.reconcile(ctx)
}
}
}
type reconcileCategory struct {
category string
dir string
}
func (r *pluginReconciler) categories(cfg config.Config) []reconcileCategory {
return []reconcileCategory{
{reconcilePlugins, cfg.PluginDir},
{reconcileModules, cfg.ModuleDir},
{reconcileSecrets, cfg.SecretEngineDir},
}
}
// reconcile re-reads the three directories against the running config and
// loads every new or changed file. It returns the outstanding problems -- a
// removal, a failed load, an unreadable directory -- joined, and reports the
// same through apply status. A nil return is a clean pass and clears the
// record's error.
func (r *pluginReconciler) reconcile(ctx context.Context) error {
r.reconcileMu.Lock()
defer r.reconcileMu.Unlock()
var problems []error
next := make(map[string]loadedPluginFile, len(r.loaded))
for _, spec := range r.categories(r.dirs()) {
problems = append(problems, r.reconcileCategory(ctx, spec, next)...)
}
for path, file := range r.loaded {
if _, ok := next[path]; ok || file.failed {
continue
}
problems = append(problems, fmt.Errorf("%s file %s was removed; it keeps running until the process restarts", file.category, path))
}
r.loaded = next
// The layering re-runs every tick, not only when a secret engine changed:
// it re-resolves provider credentials, so a provider disabled at boot for
// an unresolvable engine is re-resolved on the first tick that finds it,
// and a changed directory retargets the next tick without a restart
// (docs/prds/plugin-settings-live.md, "Triggers").
if r.relayer != nil {
if err := r.relayer(ctx); err != nil {
problems = append(problems, fmt.Errorf("re-resolve provider credentials: %w", err))
}
}
err := errors.Join(problems...)
if r.report != nil {
r.report(ctx, err)
}
return err
}
// reconcileCategory loads the new and changed files of one directory into
// next, returning the problems it met. A directory that became unreadable
// keeps the entries this category already had: it is not a removal of the
// files loaded before that.
func (r *pluginReconciler) reconcileCategory(ctx context.Context, spec reconcileCategory, next map[string]loadedPluginFile) []error {
var problems []error
paths, err := goFiles(spec.dir)
if err != nil {
problems = append(problems, fmt.Errorf("%s directory %q: %w", spec.category, spec.dir, err))
for path, file := range r.loaded {
if file.category == spec.category {
next[path] = file
}
}
return problems
}
for _, path := range paths {
hash, err := fileHash(path)
if err != nil {
problems = append(problems, fmt.Errorf("read %s: %w", path, err))
continue
}
if previous, ok := r.loaded[path]; ok && previous.hash == hash {
next[path] = previous
continue
}
if err := r.load(ctx, spec.category, path); err != nil {
problems = append(problems, fmt.Errorf("%s %s: %w", spec.category, path, err))
// Record the failed content so the same broken file is not retried
// every tick; changing it retries the load.
next[path] = loadedPluginFile{hash: hash, category: spec.category, failed: true}
continue
}
next[path] = loadedPluginFile{hash: hash, category: spec.category}
r.log.Info("runtime directory file loaded", "category", spec.category, "path", path)
}
return problems
}
func (r *pluginReconciler) load(ctx context.Context, category, path string) error {
switch category {
case reconcilePlugins:
return r.loadPlugin(ctx, path)
case reconcileModules:
return r.loadModule(path)
case reconcileSecrets:
if r.secrets == nil {
return errors.New("no secret registry to load into")
}
return r.secrets.LoadFile(path, secretEngineSymbols...)
default:
return fmt.Errorf("unknown reconcile category %q", category)
}
}
func (r *pluginReconciler) loadPlugin(ctx context.Context, path string) error {
if r.host == nil {
return errors.New("no plugin host to load into")
}
loaded, err := plugin.LoadFile(path, pluginextract.Symbols)
if err != nil {
return err
}
module, err := plugin.AdaptLegacy(loaded)
if err != nil {
return err
}
// The manifest id is the plugin's name, so an edited file that keeps its
// name replaces the running module and a new name adds one.
if r.host.Has(module.Manifest().ID) {
return r.host.Replace(ctx, module)
}
return r.host.Add(ctx, module)
}
func (r *pluginReconciler) loadModule(path string) error {
if r.modules == nil {
return errors.New("no module registry to load into")
}
kind := strings.TrimSuffix(filepath.Base(path), ".go")
if !slices.Contains(module.Kinds(), kind) {
// A .go file for a kind this build has no contract for; the boot load
// iterates the known kinds and ignores it too.
return nil
}
return r.modules.Register(kind, filepath.Dir(path))
}
// goFiles lists the .go files directly under dir, sorted. A missing directory
// reconciles to an empty set, the convention every boot loader already follows.
func goFiles(dir string) ([]string, error) {
if dir == "" {
return nil, nil
}
entries, err := os.ReadDir(dir)
if errors.Is(err, fs.ErrNotExist) {
return nil, nil
}
if err != nil {
return nil, err
}
paths := make([]string, 0, len(entries))
for _, entry := range entries {
if entry.IsDir() || !strings.HasSuffix(entry.Name(), ".go") {
continue
}
paths = append(paths, filepath.Join(dir, entry.Name()))
}
slices.Sort(paths)
return paths, nil
}
func fileHash(path string) (string, error) {
content, err := os.ReadFile(path)
if err != nil {
return "", err
}
sum := sha256.Sum256(content)
return hex.EncodeToString(sum[:]), nil
}
@@ -0,0 +1,297 @@
package archied
import (
"context"
"log/slog"
"os"
"path/filepath"
"strings"
"testing"
"github.com/samcharles93/archie-core/internal/app/controlplane"
"github.com/samcharles93/archie-core/internal/config"
"github.com/samcharles93/archie-core/internal/domain/eda/module"
"github.com/samcharles93/archie-core/internal/plugin"
"github.com/samcharles93/archie-core/internal/secret"
)
// docs/prds/plugin-settings-live.md: a plugin or secret-engine file dropped
// into the running configuration's directory loads without a restart, an
// edited file replaces what runs, and a removed file keeps running with the
// removal reported through apply status until the process restarts.
const reconcilerPluginSource = `package main
import "github.com/samcharles93/archie-core/internal/plugin"
var Plugin = plugin._Plugin{
WName: func() string { return "hello" },
WVersion: func() string { return "1.0.0" },
}
`
const reconcilerSecretSource = `package main
import "github.com/samcharles93/archie-core/internal/secret"
var Engine = secret._Engine{
WName: func() string { return "hello" },
WVersion: func() string { return "1.0.0" },
WResolve: func(key string) (string, error) { return "one:" + key, nil },
}
`
const reconcilerEditedSecretSource = `package main
import "github.com/samcharles93/archie-core/internal/secret"
var Engine = secret._Engine{
WName: func() string { return "hello" },
WVersion: func() string { return "2.0.0" },
WResolve: func(key string) (string, error) { return "two:" + key, nil },
}
`
// reconcilerHarness is one running directory set: a started plugin host, a
// fresh secret registry, an empty module registry, and the config that names
// the two directories a test writes into.
type reconcilerHarness struct {
root *plugin.Host
secrets *secret.Registry
cfg config.Config
plugins string
secretsD string
mods *module.ModuleRegistry
}
func newReconcilerHarness(t *testing.T) *reconcilerHarness {
t.Helper()
h := &reconcilerHarness{
root: plugin.NewHost(),
secrets: secret.NewRegistry(),
mods: module.New(),
plugins: t.TempDir(),
secretsD: t.TempDir(),
}
if err := h.root.Start(t.Context()); err != nil {
t.Fatalf("start host: %v", err)
}
h.cfg = config.Config{PluginDir: h.plugins, SecretEngineDir: h.secretsD}
return h
}
func (h *reconcilerHarness) reconciler() *pluginReconciler {
return newPluginReconciler(slog.New(slog.DiscardHandler), pluginReconcileTargets{
host: h.root,
secrets: h.secrets,
modules: h.mods,
dirs: func() config.Config { return h.cfg },
})
}
func writeReconcilerFile(t *testing.T, path, source string) {
t.Helper()
if err := os.WriteFile(path, []byte(source), 0o644); err != nil {
t.Fatal(err)
}
}
func TestPluginReconcilerLoadsADroppedPluginAndSecretEngine(t *testing.T) {
h := newReconcilerHarness(t)
r := h.reconciler()
// Boot loaded the (empty) directories; seed is what the daemon does after
// its own boot load so the first tick does not re-load the boot set.
r.seed()
writeReconcilerFile(t, filepath.Join(h.plugins, "hello.go"), reconcilerPluginSource)
writeReconcilerFile(t, filepath.Join(h.secretsD, "hello.go"), reconcilerSecretSource)
if err := r.reconcile(t.Context()); err != nil {
t.Fatalf("reconcile: %v", err)
}
if !h.root.Has("hello") {
t.Error("dropped plugin was not loaded in place")
}
engine, ok := h.secrets.Get("hello")
if !ok {
t.Fatal("dropped secret engine was not loaded in place")
}
value, err := engine.Resolve("k")
if err != nil {
t.Fatalf("Resolve: %v", err)
}
if value != "one:k" {
t.Errorf("Resolve() = %q, want one:k", value)
}
}
func TestPluginReconcilerReplacesAnEditedSecretEngine(t *testing.T) {
h := newReconcilerHarness(t)
r := h.reconciler()
r.seed()
path := filepath.Join(h.secretsD, "hello.go")
writeReconcilerFile(t, path, reconcilerSecretSource)
if err := r.reconcile(t.Context()); err != nil {
t.Fatalf("first reconcile: %v", err)
}
writeReconcilerFile(t, path, reconcilerEditedSecretSource)
if err := r.reconcile(t.Context()); err != nil {
t.Fatalf("second reconcile: %v", err)
}
engine, ok := h.secrets.Get("hello")
if !ok {
t.Fatal("edited secret engine disappeared")
}
value, err := engine.Resolve("k")
if err != nil {
t.Fatalf("Resolve: %v", err)
}
if value != "two:k" {
t.Errorf("Resolve() after edit = %q, want the replaced two:k", value)
}
}
func TestPluginReconcilerReportsARemovedFileThatKeepsRunning(t *testing.T) {
h := newReconcilerHarness(t)
r := h.reconciler()
r.seed()
pluginPath := filepath.Join(h.plugins, "hello.go")
secretPath := filepath.Join(h.secretsD, "hello.go")
writeReconcilerFile(t, pluginPath, reconcilerPluginSource)
writeReconcilerFile(t, secretPath, reconcilerSecretSource)
if err := r.reconcile(t.Context()); err != nil {
t.Fatalf("first reconcile: %v", err)
}
if err := os.Remove(pluginPath); err != nil {
t.Fatal(err)
}
if err := os.Remove(secretPath); err != nil {
t.Fatal(err)
}
err := r.reconcile(t.Context())
if err == nil {
t.Fatal("a reconcile after two removals reported no problem")
}
for _, want := range []string{"hello.go", "removed", "restart"} {
if !strings.Contains(err.Error(), want) {
t.Errorf("removal report %q does not mention %q", err, want)
}
}
// The running code is not unloaded: Yaegi has no unload path, so both the
// module and the engine keep serving until the process restarts.
if !h.root.Has("hello") {
t.Error("a removed plugin stopped running")
}
if _, ok := h.secrets.Get("hello"); !ok {
t.Error("a removed secret engine was unregistered")
}
// Re-adding the file clears the removal: the replacement takes over and no
// old interpreter is left serving the path.
writeReconcilerFile(t, pluginPath, reconcilerPluginSource)
writeReconcilerFile(t, secretPath, reconcilerSecretSource)
if err := r.reconcile(t.Context()); err != nil {
t.Fatalf("reconcile after re-adding the files: %v", err)
}
}
func TestPluginSettingsLiveUpdateLoadsTheNewDirectory(t *testing.T) {
dir := t.TempDir()
writeReconcilerFile(t, filepath.Join(dir, "hello.go"), reconcilerPluginSource)
running := fileConfig()
b, status := newLiveApplyBoot(t, running)
resources := databaseOwnedResources()
resources[controlplane.PluginSettingsKind] = map[string]any{
"plugin_dir": dir, "module_dir": "/db/modules", "secret_engine_dir": "/db/secrets", "skills_dir": "",
}
b.controlPlane = controlplane.NewRPCClient(&controlPlaneStub{values: resources})
b.secrets = secret.NewRegistry()
host := plugin.NewHost()
if err := host.Start(t.Context()); err != nil {
t.Fatalf("start host: %v", err)
}
b.pluginReconciler = newPluginReconciler(b.log, pluginReconcileTargets{
host: host,
secrets: b.secrets,
modules: module.New(),
dirs: func() config.Config { return b.cfgHolder.Get() },
})
b.pluginReconciler.seed()
b.applyRuntimeResourceUpdate(t.Context(), controlplane.PluginSettingsKind, controlplane.AppliedResource{Version: 2})
if got := b.cfgHolder.Get().PluginDir; got != dir {
t.Fatalf("running plugin dir = %q, want the stored %q", got, dir)
}
if !host.Has("hello") {
t.Fatal("a stored plugin_dir change did not load its plugin without a restart")
}
last, ok := status.last(controlplane.PluginSettingsKind)
if !ok {
t.Fatal("the live plugin-settings update reported no apply status")
}
if last.Error != "" {
t.Errorf("apply status error = %q, want none for a live directory load", last.Error)
}
}
func TestPluginSettingsSkillsDirChangeIsRefusedOnTheLivePath(t *testing.T) {
running := fileConfig()
b, status := newLiveApplyBoot(t, running)
// The stored document's skills_dir (/db/skills) differs from the running
// empty value, and skills_dir has no live consumer.
b.controlPlane = controlplane.NewRPCClient(&controlPlaneStub{values: databaseOwnedResources()})
b.applyRuntimeResourceUpdate(t.Context(), controlplane.PluginSettingsKind, controlplane.AppliedResource{Version: 2})
if got := b.cfgHolder.Get().PluginDir; got != fileConfig().PluginDir {
t.Errorf("a refused plugin-settings document published plugin_dir %q, want the file's %q", got, fileConfig().PluginDir)
}
last, ok := status.last(controlplane.PluginSettingsKind)
if !ok {
t.Fatal("the refused plugin-settings update reported no apply status")
}
if !strings.Contains(last.Error, "skills_dir") {
t.Errorf("apply status error = %q, want it to name skills_dir", last.Error)
}
if last.AppliedVersion != 2 {
t.Errorf("apply status version = %d, want the refused version 2", last.AppliedVersion)
}
}
func TestPluginReconcilerReportsTheApplyStatusTruth(t *testing.T) {
h := newReconcilerHarness(t)
var reported []error
reported = nil
r := newPluginReconciler(slog.New(slog.DiscardHandler), pluginReconcileTargets{
host: h.root,
secrets: h.secrets,
modules: h.mods,
dirs: func() config.Config { return h.cfg },
report: func(_ context.Context, err error) { reported = append(reported, err) },
})
r.seed()
writeReconcilerFile(t, filepath.Join(h.secretsD, "hello.go"), reconcilerSecretSource)
if err := r.reconcile(t.Context()); err != nil {
t.Fatalf("clean reconcile: %v", err)
}
if len(reported) != 1 || reported[0] != nil {
t.Fatalf("clean reconcile reported %v, want one nil (a live add is applied)", reported)
}
if err := os.Remove(filepath.Join(h.secretsD, "hello.go")); err != nil {
t.Fatal(err)
}
if err := r.reconcile(t.Context()); err == nil {
t.Fatal("removal reconcile reported no error")
}
if len(reported) != 2 || reported[1] == nil {
t.Fatalf("removal reconcile reported %v, want the removal as the second record", reported)
}
}
+11 -1
View File
@@ -5,6 +5,7 @@ import (
"fmt"
"log/slog"
"os"
"reflect"
"strings"
"github.com/traefik/yaegi/stdlib/unrestricted"
@@ -15,6 +16,15 @@ import (
"github.com/samcharles93/archie-core/internal/secret/secretextract"
)
// secretEngineSymbols is the Yaegi symbol set every secret-engine load uses:
// the generated interface bridge, the host-side CLI helper, and unrestricted
// symbols for parity with skillscript interpretation. The boot load and a
// live directory reconciliation share it so an engine always loads the same
// way.
var secretEngineSymbols = []map[string]map[string]reflect.Value{
secretextract.Symbols, enginehost.Symbols, unrestricted.Symbols,
}
func configuredSecretRegistry(cfg *config.Config, log *slog.Logger) (*secret.Registry, error) {
registry := secret.NewRegistry()
if cfg.SecretEngineDir != "" {
@@ -24,7 +34,7 @@ func configuredSecretRegistry(cfg *config.Config, log *slog.Logger) (*secret.Reg
// host-side helper) because yaegi's interpreted os/exec cannot
// pass an environment to child processes; unrestricted symbols
// are included for parity with skillscript interpretation.
loaded, err := registry.LoadDir(cfg.SecretEngineDir, secretextract.Symbols, enginehost.Symbols, unrestricted.Symbols)
loaded, err := registry.LoadDir(cfg.SecretEngineDir, secretEngineSymbols...)
if err != nil {
return nil, fmt.Errorf("load secret engines from %q: %w", cfg.SecretEngineDir, err)
}
+19 -13
View File
@@ -135,11 +135,13 @@ var reloadableFields = map[string]bool{
// TestReloadableFieldsCoverForTaskSnapshot.
"Label": true,
"BotUser": true,
// Worked example of the every-consumer criterion above: the sole
// consumer is the webui maxRetriesFor handler (api_tasks.go:380,383),
// which re-reads per request; the daemon never reads MaxRetries. The
// per-task half of this list is pinned mechanically; this webui-only
// entry is pinned by TestChangedNonReloadableFields.
// MaxRetries has two consumers, both re-reading it: the webui
// maxRetriesFor handler (api_tasks.go:380,383) reads it per request, and
// ForTask carries it into every dispatched TaskConfig, where the
// container's remediation round cap reads it (StageRemediationRoundCap).
// The per-task half of this list is pinned mechanically by
// TestReloadableFieldsCoverForTaskSnapshot; the webui-only entry is pinned
// by TestChangedNonReloadableFields.
"MaxRetries": true,
// Tools.Policy (MaxResultChars/SpillDir) is carried into TaskConfig by
// ForTask (config.go) and applied fresh per dispatch via
@@ -156,17 +158,21 @@ var reloadableFields = map[string]bool{
// "Selection": a Kit profile applies without a restart. Credentials is the
// same shape one layer down: kitrun.Launcher reads Config.Get().Containers.
// Credentials fresh on every Launch (CredentialBindingsKind), never a slice
// captured once when the launcher was built. Everything else in
// Containers (Image, MaxConcurrency, MaxUptime, PullPolicy, Network) is
// frozen in the startup-built container pool -- the dispatchers re-read
// MaxConcurrency but the pool captures it at construction
// (container/pool.go:94,163), so a change only partially applies and must
// warn requires-restart. Forge.Host is carried into TaskContext by ForTask
// captured once when the launcher was built. Image, MaxConcurrency, MaxUptime,
// PullPolicy and Network are re-read per acquire by the container pool and
// MaxConcurrency also resizes the running dispatcher on publish
// (archie-core-zfb0.2), so a reload applies them without a restart.
// RegistryAuth stays frozen in the startup-built pool: it is file-owned and
// never part of the container-runtime-policies document, so a changed value
// must warn requires-restart. Forge.Host is carried into TaskContext by ForTask
// (display/link building only); the forge client itself is startup-built, so
// Type/Token/TokenEnv stay requires-restart.
var reloadableSubFields = map[string]map[string]bool{
"Containers": {"VolumeTTL": true, "Profiles": true, "Credentials": true},
"Forge": {"Host": true},
"Containers": {
"VolumeTTL": true, "Profiles": true, "Credentials": true,
"Image": true, "MaxConcurrency": true, "MaxUptime": true, "PullPolicy": true, "Network": true,
},
"Forge": {"Host": true},
// Policy is carried into TaskConfig by ForTask (config.go); MCPServers,
// WebFetch and Minimax are not and stay requires-restart.
"Tools": {"Policy": true},
+12 -9
View File
@@ -198,28 +198,31 @@ func TestChangedNonReloadableFields(t *testing.T) {
t.Fatalf("reloadable-only: got %v, want []", got)
}
// Containers sub-field granularity: Image change warns, MaxConcurrency
// and VolumeTTL changes do not.
// Containers sub-field granularity: the fields the pool re-reads per
// acquire, and the concurrency cap the dispatcher resizes on, are
// reloadable; RegistryAuth stays frozen in the startup-built pool.
c1 := base
c1.Containers = config.ContainerConfig{Image: "img1", MaxConcurrency: 2, VolumeTTL: config.Duration(24 * time.Hour)}
c2 := c1
c2.Containers.Image = "img2"
if got := changedNonReloadableFields(c1, c2); len(got) != 1 || got[0] != "Containers.Image" {
t.Fatalf("Image change: got %v, want [Containers.Image]", got)
if got := changedNonReloadableFields(c1, c2); len(got) != 0 {
t.Fatalf("Image change: got %v, want [] (the pool reads it per acquire)", got)
}
c3 := c1
c3.Containers.MaxConcurrency = 4
// MaxConcurrency is NOT reloadable: the dispatchers re-read it but the
// startup-built container pool captures it (container/pool.go:94), so
// a change only partially applies and must warn requires-restart.
if got := changedNonReloadableFields(c1, c3); len(got) != 1 || got[0] != "Containers.MaxConcurrency" {
t.Fatalf("MaxConcurrency change: got %v, want [Containers.MaxConcurrency]", got)
if got := changedNonReloadableFields(c1, c3); len(got) != 0 {
t.Fatalf("MaxConcurrency change: got %v, want [] (the pool reads it and the dispatcher resizes)", got)
}
c4 := c1
c4.Containers.VolumeTTL = config.Duration(48 * time.Hour)
if got := changedNonReloadableFields(c1, c4); len(got) != 0 {
t.Fatalf("VolumeTTL change: got %v, want []", got)
}
c5 := c1
c5.Containers.RegistryAuth = config.SecretRef{Engine: "env", Key: "REGISTRY_AUTH"}
if got := changedNonReloadableFields(c1, c5); len(got) != 1 || got[0] != "Containers.RegistryAuth" {
t.Fatalf("RegistryAuth change: got %v, want [Containers.RegistryAuth] (the pool resolves it at boot)", got)
}
// NATS is not reloadable: a URL change warns (the daemon's client is
// startup-built; containers get the frozen ConnectedNATS).
@@ -15,6 +15,7 @@ import (
"github.com/samcharles93/archie-core/internal/config"
"github.com/samcharles93/archie-core/internal/infrastructure/postgres/pgtest"
"github.com/samcharles93/archie-core/internal/secret"
)
// startupGrace is how long RunStateStore is given to either refuse the config
@@ -46,7 +47,7 @@ func TestExternalServicesDoNotRequireBridgeNetwork(t *testing.T) {
cfg.NATS.Mode = config.NATSModeExternal
cfg.Containers.Network = "host"
cfg.Services = config.Services{config.ServiceNameState: {Target: "state.example.test:9090"}}
pool, _, cleanup := startContainers(t.Context(), cfg, slog.New(slog.DiscardHandler))
pool, _, cleanup := startContainers(t.Context(), config.NewHolder(cfg), secret.NewRegistry(), slog.New(slog.DiscardHandler))
defer cleanup()
if pool == nil {
t.Fatal("external services on host networking must construct a worker pool")
@@ -24,9 +24,9 @@ import (
type testTaskActor struct{ b *boot }
func (a testTaskActor) ApplyChatTaskAction(
ctx context.Context, identity *string, actor taskactions.Actor, taskID int64, action taskstate.Action,
ctx context.Context, identity *string, actor taskactions.Actor, taskID int64, action taskstate.Action, res taskactions.ActionPayload,
) (gateway.TaskActionResult, error) {
if err := a.b.taskActions().Apply(ctx, identity, actor, taskID, action); err != nil {
if err := a.b.taskActions().Apply(ctx, identity, actor, taskID, action, res); err != nil {
return gateway.TaskActionResult{}, err
}
return gateway.TaskActionResult{
@@ -84,7 +84,7 @@ func TestChatTaskActorAdapterCrossIdentityRefused(t *testing.T) {
for _, tc := range tests {
t.Run(tc.name, func(t *testing.T) {
_, err := adapter.ApplyChatTaskAction(ctx, &tc.actorIdentity, taskactions.Actor{}, task.ID, tc.action)
_, err := adapter.ApplyChatTaskAction(ctx, &tc.actorIdentity, taskactions.Actor{}, task.ID, tc.action, taskactions.ActionPayload{})
if err == nil {
t.Fatalf("ApplyChatTaskAction() allowed cross-identity action %s by %q on task owned by %q",
tc.action, tc.actorIdentity, task.Identity)
@@ -152,7 +152,7 @@ func TestChatTaskActorAdapterRefusesDisallowedStateAction(t *testing.T) {
}
taskstatetest.Seed(ctx, t, st, task.ID, workflow.StatusQueued, tc.initialStatus, "setup")
_, err = adapter.ApplyChatTaskAction(ctx, new("archie"), taskactions.Actor{}, task.ID, tc.attemptAction)
_, err = adapter.ApplyChatTaskAction(ctx, new("archie"), taskactions.Actor{}, task.ID, tc.attemptAction, taskactions.ActionPayload{})
if err == nil {
t.Fatalf("expected error attempting %s on %s task, got nil", tc.attemptAction, tc.initialStatus)
}
@@ -182,7 +182,7 @@ func TestChatTaskActorAdapterTaskNotFound(t *testing.T) {
adapter := newChatTaskActorForTest(t, st, config.Config{})
_, err := adapter.ApplyChatTaskAction(ctx, new("archie"), taskactions.Actor{}, 999999, taskstate.ActionAbandon)
_, err := adapter.ApplyChatTaskAction(ctx, new("archie"), taskactions.Actor{}, 999999, taskstate.ActionAbandon, taskactions.ActionPayload{})
if err == nil {
t.Fatal("expected error for non-existent task, got nil")
}
@@ -257,7 +257,7 @@ func TestChatTaskActorAdapterAppliesActionsToStore(t *testing.T) {
}
taskstatetest.Seed(ctx, t, st, task.ID, workflow.StatusQueued, tc.initialStatus, "setup")
result, err := adapter.ApplyChatTaskAction(ctx, new("archie"), taskactions.Actor{}, task.ID, tc.action)
result, err := adapter.ApplyChatTaskAction(ctx, new("archie"), taskactions.Actor{}, task.ID, tc.action, taskactions.ActionPayload{})
if err != nil {
t.Fatalf("ApplyChatTaskAction(%s) error = %v", tc.action, err)
}
@@ -292,7 +292,7 @@ func TestChatTaskActorAdapterNilContract(t *testing.T) {
ctx := context.Background()
adapter := chatTaskActorAdapter{contract: nil}
_, err := adapter.ApplyChatTaskAction(ctx, new("archie"), taskactions.Actor{}, 1, taskstate.ActionAbandon)
_, err := adapter.ApplyChatTaskAction(ctx, new("archie"), taskactions.Actor{}, 1, taskstate.ActionAbandon, taskactions.ActionPayload{})
if err == nil {
t.Fatal("expected error for nil contract, got nil")
}
+10 -2
View File
@@ -11,9 +11,12 @@ import (
"github.com/samcharles93/archie-core/internal/config"
"github.com/samcharles93/archie-core/internal/domain/storecontract"
"github.com/samcharles93/archie-core/internal/domain/taskactions"
"github.com/samcharles93/archie-core/internal/domain/workflow"
"github.com/samcharles93/archie-core/internal/gateway"
"github.com/samcharles93/archie-core/internal/infrastructure/postgres/pgstore"
taskactionstore "github.com/samcharles93/archie-core/internal/infrastructure/taskactions"
"github.com/samcharles93/archie-core/internal/taskstate"
"github.com/samcharles93/archie-core/internal/taskstate/taskstatetest"
"github.com/samcharles93/archie-core/internal/webui"
)
@@ -80,9 +83,14 @@ func runChatAction(t *testing.T, ctx context.Context, st storecontract.TaskStore
if !ok {
t.Fatal("the task store cannot cancel executions")
}
// Chat's /approve reaches the daemon's one task-action service
// (docs/prds/pr-review-operator-response.md, Decision 1).
actions := taskactionstore.NewService(taskactionstore.Store{TaskStore: st}, nil, nil, nil, nil, nil, nil, nil)
controller := gateway.NewStoreTaskController(chatTaskControllerAdapter{
taskByID: st.TaskByID,
requeue: st.Requeue,
taskByID: st.TaskByID,
approve: func(ctx context.Context, scope *string, actor taskactions.Actor, taskID int64, res taskactions.ActionPayload) error {
return actions.Apply(ctx, scope, actor, taskID, taskstate.ActionApprove, res)
},
cancelExecution: canceller.CancelExecution,
})
var err error
@@ -0,0 +1,337 @@
package archied
import (
"context"
"errors"
"strings"
"sync"
"testing"
"github.com/samcharles93/archie-core/internal/app/controlplane"
"github.com/samcharles93/archie-core/internal/config"
"github.com/samcharles93/archie-core/internal/plugin"
"github.com/samcharles93/archie-core/internal/secret"
"github.com/samcharles93/archie-core/internal/tools"
toolprovider "github.com/samcharles93/archie-core/internal/tools/provider"
"github.com/samcharles93/archie-core/internal/tools/webfetch"
)
// fakeMCPEngine stands in for a configured MCP server so the live reconciliation
// can be driven without spawning a process. It records each lifecycle call on
// the shared recorder, keyed by configured server name.
type fakeMCPEngine struct {
name string
tools []tools.ToolEntry
rec *mcpEngineRecorder
startErr error
}
func (e *fakeMCPEngine) Manifest() plugin.Manifest {
return plugin.Manifest{
ID: "mcp." + e.name,
Name: e.name,
Version: "1.0.0",
APIVersion: plugin.HostAPIVersion,
Capabilities: []plugin.CapabilityKind{"tools"},
}
}
func (e *fakeMCPEngine) Start(context.Context) error {
e.rec.recordStart(e.name)
return e.startErr
}
func (e *fakeMCPEngine) Stop(context.Context) error {
e.rec.recordStop(e.name)
return nil
}
func (e *fakeMCPEngine) Discover(context.Context) ([]tools.ToolEntry, error) {
return append([]tools.ToolEntry(nil), e.tools...), nil
}
// mcpEngineRecorder records what the live reconciliation did to each server and
// hands the test a factory boot.mcpProvider can use.
type mcpEngineRecorder struct {
mu sync.Mutex
builds map[string]int
starts map[string]int
stops map[string]int
failNext map[string]error
seen map[string]config.MCPServer
}
func newMCPEngineRecorder() *mcpEngineRecorder {
return &mcpEngineRecorder{
builds: map[string]int{},
starts: map[string]int{},
stops: map[string]int{},
failNext: map[string]error{},
seen: map[string]config.MCPServer{},
}
}
func (r *mcpEngineRecorder) factory() func(config.MCPServer) (toolprovider.Engine, error) {
return func(srv config.MCPServer) (toolprovider.Engine, error) {
name := strings.TrimSpace(srv.Name)
r.mu.Lock()
r.builds[name]++
r.seen[name] = srv
err := r.failNext[name]
delete(r.failNext, name)
r.mu.Unlock()
return &fakeMCPEngine{
name: name,
rec: r,
startErr: err,
tools: []tools.ToolEntry{{
Name: "mcp." + name + ".tool",
Toolset: "mcp",
Handler: func(context.Context, map[string]any) (any, error) { return "ok", nil },
}},
}, nil
}
}
func (r *mcpEngineRecorder) recordStart(name string) {
r.mu.Lock()
defer r.mu.Unlock()
r.starts[name]++
}
func (r *mcpEngineRecorder) recordStop(name string) {
r.mu.Lock()
defer r.mu.Unlock()
r.stops[name]++
}
func (r *mcpEngineRecorder) failOnNextStart(name string, err error) {
r.mu.Lock()
defer r.mu.Unlock()
r.failNext[name] = err
}
func (r *mcpEngineRecorder) count(counts map[string]int, name string) int {
r.mu.Lock()
defer r.mu.Unlock()
return counts[name]
}
// toolSettingsDocument renders a stored tool-settings document carrying the
// given MCP servers and an enabled web_fetch.
func toolSettingsDocument(servers ...map[string]any) map[string]any {
return map[string]any{
"mcp_servers": servers,
"policy": map[string]any{},
"web_fetch": map[string]any{},
"minimax": map[string]any{},
}
}
func mcpServerDoc(name string, parallel bool) map[string]any {
return map[string]any{
"name": name,
"transport": "stdio",
"command": "run-server",
"parallel_tool_calls": parallel,
}
}
// newMCPReconcileBoot builds a boot whose provider registry is running, so the
// live reconciliation acts on a real registry and a real tool index.
func newMCPReconcileBoot(t *testing.T, base config.Config, stub *resourceWatchStub) (*boot, *applyStatusRecorder, *mcpEngineRecorder) {
t.Helper()
b, status := newLiveApplyBoot(t, base)
b.controlPlane = controlplane.NewRPCClient(stub)
b.secrets = secret.NewRegistry()
b.toolReg = tools.NewRegistry()
b.providerRegistry = toolprovider.NewRegistry(b.toolReg)
if err := b.providerRegistry.Start(t.Context()); err != nil {
t.Fatalf("start provider registry: %v", err)
}
b.mcpApplied = map[string]appliedMCPServer{}
rec := newMCPEngineRecorder()
b.mcpProvider = rec.factory()
return b, status, rec
}
// applyToolSettings stores the document and runs the live apply the watch
// would run, the seam the daemon meets a stored tool-settings change through.
func applyToolSettings(t *testing.T, b *boot, stub *resourceWatchStub, version int64, document map[string]any) {
t.Helper()
if stub.values == nil {
stub.values = map[string]any{}
}
stub.values[controlplane.ToolSettingsKind] = document
if stub.storedVersion == nil {
stub.storedVersion = map[string]int64{}
}
stub.storedVersion[controlplane.ToolSettingsKind] = version
b.applyRuntimeResourceUpdate(t.Context(), controlplane.ToolSettingsKind, controlplane.AppliedResource{Version: version})
}
func TestLiveToolSettingsConnectsAnAddedServer(t *testing.T) {
stub := &resourceWatchStub{values: databaseOwnedResources(), version: 10, storedVersion: map[string]int64{}, opens: map[string]int{}}
b, _, rec := newMCPReconcileBoot(t, fileConfig(), stub)
applyToolSettings(t, b, stub, 11, toolSettingsDocument(mcpServerDoc("alpha", false)))
if got := rec.count(rec.starts, "alpha"); got != 1 {
t.Errorf("alpha start count = %d, want 1", got)
}
if _, ok := b.toolReg.Get("mcp.alpha.tool"); !ok {
t.Error("the added server's tool was not indexed")
}
if !b.providerRegistry.Has("mcp.alpha") {
t.Error("the added server is not registered")
}
servers := b.cfgHolder.Get().Tools.MCPServers
if len(servers) != 1 || servers[0].Name != "alpha" {
t.Errorf("running config servers = %+v, want the added alpha", servers)
}
}
func TestLiveToolSettingsLeavesAnUnchangedServerUntouched(t *testing.T) {
stub := &resourceWatchStub{values: databaseOwnedResources(), version: 10, storedVersion: map[string]int64{}, opens: map[string]int{}}
b, _, rec := newMCPReconcileBoot(t, fileConfig(), stub)
applyToolSettings(t, b, stub, 11, toolSettingsDocument(mcpServerDoc("alpha", false)))
applyToolSettings(t, b, stub, 12, toolSettingsDocument(mcpServerDoc("alpha", false), mcpServerDoc("beta", false)))
if got := rec.count(rec.builds, "alpha"); got != 1 {
t.Errorf("alpha was rebuilt on an unchanged update: builds = %d, want 1", got)
}
if got := rec.count(rec.starts, "alpha"); got != 1 {
t.Errorf("alpha was reconnected on an unchanged update: starts = %d, want 1", got)
}
if got := rec.count(rec.stops, "alpha"); got != 0 {
t.Errorf("alpha was stopped on an unchanged update: stops = %d, want 0", got)
}
if got := rec.count(rec.starts, "beta"); got != 1 {
t.Errorf("beta start count = %d, want 1", got)
}
}
func TestLiveToolSettingsReconnectsAChangedServer(t *testing.T) {
stub := &resourceWatchStub{values: databaseOwnedResources(), version: 10, storedVersion: map[string]int64{}, opens: map[string]int{}}
b, _, rec := newMCPReconcileBoot(t, fileConfig(), stub)
applyToolSettings(t, b, stub, 11, toolSettingsDocument(mcpServerDoc("alpha", false)))
applyToolSettings(t, b, stub, 12, toolSettingsDocument(mcpServerDoc("alpha", true)))
if got := rec.count(rec.stops, "alpha"); got != 1 {
t.Errorf("the changed server was not disconnected: stops = %d, want 1", got)
}
if got := rec.count(rec.starts, "alpha"); got != 2 {
t.Errorf("the changed server was not reconnected: starts = %d, want 2", got)
}
if _, ok := b.toolReg.Get("mcp.alpha.tool"); !ok {
t.Error("the reconnected server's tool left the index")
}
}
func TestLiveToolSettingsDisconnectsARemovedServer(t *testing.T) {
stub := &resourceWatchStub{values: databaseOwnedResources(), version: 10, storedVersion: map[string]int64{}, opens: map[string]int{}}
b, _, rec := newMCPReconcileBoot(t, fileConfig(), stub)
applyToolSettings(t, b, stub, 11, toolSettingsDocument(mcpServerDoc("alpha", false), mcpServerDoc("beta", false)))
applyToolSettings(t, b, stub, 12, toolSettingsDocument(mcpServerDoc("beta", false)))
if got := rec.count(rec.stops, "alpha"); got != 1 {
t.Errorf("the removed server was not disconnected: stops = %d, want 1", got)
}
if b.providerRegistry.Has("mcp.alpha") {
t.Error("the removed server stayed registered")
}
if _, ok := b.toolReg.Get("mcp.alpha.tool"); ok {
t.Error("the removed server's tool stayed indexed")
}
if got := rec.count(rec.stops, "beta"); got != 0 {
t.Errorf("the surviving server was disturbed: stops = %d, want 0", got)
}
}
// TestLiveToolSettingsKeepsTheRunningServerWhenAChangeCannotStart is the path
// where the change must NOT apply: a changed server whose replacement fails to
// start is rolled back to the one that was running, and the refusal reaches
// apply status rather than the kind silently reporting success.
func TestLiveToolSettingsKeepsTheRunningServerWhenAChangeCannotStart(t *testing.T) {
stub := &resourceWatchStub{values: databaseOwnedResources(), version: 10, storedVersion: map[string]int64{}, opens: map[string]int{}}
b, status, rec := newMCPReconcileBoot(t, fileConfig(), stub)
applyToolSettings(t, b, stub, 11, toolSettingsDocument(mcpServerDoc("alpha", false)))
rec.failOnNextStart("alpha", errors.New("handshake refused"))
applyToolSettings(t, b, stub, 12, toolSettingsDocument(mcpServerDoc("alpha", true)))
// Two stops: the running server is disconnected for the replacement, and
// the replacement's own half-start is cleaned up when its handshake fails.
if got := rec.count(rec.stops, "alpha"); got != 2 {
t.Errorf("stops = %d, want 2 (the replaced server and the failed replacement's cleanup)", got)
}
if got := rec.count(rec.starts, "alpha"); got != 3 {
t.Errorf("starts = %d, want 3 (initial, refused replacement, rollback restart)", got)
}
if _, ok := b.toolReg.Get("mcp.alpha.tool"); !ok {
t.Error("the refused change left the running server's tool out of the index")
}
if !b.providerRegistry.Has("mcp.alpha") {
t.Error("the refused change left the server unregistered")
}
last, ok := status.last(controlplane.ToolSettingsKind)
if !ok {
t.Fatal("the refused change reported no apply status")
}
if !strings.Contains(last.Error, "alpha") {
t.Errorf("apply status error = %q, want it to name the refused server", last.Error)
}
}
// TestLiveToolSettingsRebuildsTheWebFetchEntry: web_fetch captured its config
// at construction, so a live change rebuilds the entry rather than leaving the
// boot value in the model's tool list.
func TestLiveToolSettingsRebuildsTheWebFetchEntry(t *testing.T) {
stub := &resourceWatchStub{values: databaseOwnedResources(), version: 10, storedVersion: map[string]int64{}, opens: map[string]int{}}
b, _, _ := newMCPReconcileBoot(t, fileConfig(), stub)
disabled := toolSettingsDocument()
disabled["web_fetch"] = map[string]any{"enabled": false}
applyToolSettings(t, b, stub, 11, disabled)
if _, ok := b.toolReg.Get(webfetch.ToolName); ok {
t.Error("web_fetch stayed registered after the stored document disabled it")
}
applyToolSettings(t, b, stub, 12, toolSettingsDocument())
if _, ok := b.toolReg.Get(webfetch.ToolName); !ok {
t.Error("web_fetch was not registered after the stored document re-enabled it")
}
}
// TestLiveToolSettingsKeepsAnUnprojectedMCPHeader: the stored document cannot
// carry secret-bearing headers, so the layering restores them from the running
// base (controlplane.runtimeToolConfigFrom). The reconciliation must build the
// reconnected engine from that layered server, not from the document alone, or
// an authentication header disappears on the first live edit.
func TestLiveToolSettingsKeepsAnUnprojectedMCPHeader(t *testing.T) {
base := fileConfig()
base.Tools.MCPServers = []config.MCPServer{{
Name: "alpha",
Transport: "http",
URL: "https://alpha.example/mcp",
Headers: map[string]string{"Authorization": "Bearer secret"},
}}
stub := &resourceWatchStub{values: databaseOwnedResources(), version: 10, storedVersion: map[string]int64{}, opens: map[string]int{}}
b, _, rec := newMCPReconcileBoot(t, base, stub)
changed := mcpServerDoc("alpha", true)
changed["transport"] = "http"
changed["url"] = "https://alpha.example/mcp"
delete(changed, "command")
applyToolSettings(t, b, stub, 11, toolSettingsDocument(changed))
rec.mu.Lock()
seen := rec.seen["alpha"]
rec.mu.Unlock()
if got := seen.Headers["Authorization"]; got != "Bearer secret" {
t.Errorf("reconnected server headers = %v, want the unprojected Authorization header restored", seen.Headers)
}
}
+39
View File
@@ -0,0 +1,39 @@
package archied
import (
"log/slog"
"testing"
"github.com/samcharles93/archie-core/internal/config"
"github.com/samcharles93/archie-core/internal/infrastructure/transcription"
"github.com/samcharles93/archie-core/internal/secret"
)
// TestSetupTranscriberBuildsOnTheModelOwningSide pins where the voice
// capability is constructed: beside the chat runtime, from the process's own
// [models].transcription and [providers.*]. The Messaging Service carries the
// audio bytes across the inbound wire but no longer builds this client.
func TestSetupTranscriberBuildsOnTheModelOwningSide(t *testing.T) {
log := slog.New(slog.DiscardHandler)
t.Run("configured role wires a transcriber", func(t *testing.T) {
b := &boot{secrets: secret.NewRegistry()}
b.setupTranscriber(config.Config{
Models: map[string]string{transcription.Role: "local/whisper-large-v3"},
Providers: map[string]config.Provider{
"local": {Class: "openai-compatible", BaseURL: "http://127.0.0.1:8080/v1"},
},
}, log)
if b.transcriber == nil {
t.Fatal("transcriber = nil, want a client built from [models].transcription")
}
})
t.Run("absent role degrades to nil", func(t *testing.T) {
b := &boot{secrets: secret.NewRegistry()}
b.setupTranscriber(config.Config{}, log)
if b.transcriber != nil {
t.Fatalf("transcriber = %v, want nil when no role is configured", b.transcriber)
}
})
}
@@ -15,7 +15,7 @@ import (
// channel. Only what an operator may see: no configuration value ever enters a
// descriptor, so the manager cannot become a second source for a token or a
// credential.
func channelDescriptors(instances []channelInstance) []status.Descriptor {
func channelDescriptors(instances []*channelInstance) []status.Descriptor {
descriptors := make([]status.Descriptor, 0, len(instances))
for _, instance := range instances {
descriptors = append(descriptors, status.Descriptor{
@@ -34,8 +34,8 @@ func channelDescriptors(instances []channelInstance) []status.Descriptor {
// the single source for both the capability a descriptor declares and the action
// a reload performs; deriving them separately is how a dashboard ends up offering
// a button that reports success and changes nothing (archie-core-np7l).
func reloadable(instance channelInstance) bool {
gateway, ok := instance.channel.(*telegram.Gateway)
func reloadable(instance *channelInstance) bool {
gateway, ok := instance.current().(*telegram.Gateway)
return ok && gateway.Reload != nil
}
@@ -53,7 +53,7 @@ func (s *Service) ReloadChannel(_ context.Context, id string) error {
if instance.name != id {
continue
}
gateway, ok := instance.channel.(*telegram.Gateway)
gateway, ok := instance.current().(*telegram.Gateway)
if !ok || gateway.Reload == nil {
return fmt.Errorf("messaging: channel %q does not support reload", id)
}
@@ -144,7 +144,7 @@ func TestChannelDescriptorsDeclareCapabilities(t *testing.T) {
// "telegram" correctly declares nothing.
withReload := telegram.New("token-123", []int64{1}, slog.New(slog.DiscardHandler))
withReload.Reload = func(*telegram.Gateway) error { return nil }
descriptors := channelDescriptors([]channelInstance{
descriptors := channelDescriptors([]*channelInstance{
{name: "telegram", channel: withReload},
{name: "email", channel: fakeChannel{}},
{name: "webhook", channel: fakeChannel{}},
+169 -68
View File
@@ -7,6 +7,8 @@ import (
"net"
"strconv"
"sync"
"sync/atomic"
"time"
"github.com/samcharles93/archie-core/internal/channels"
"github.com/samcharles93/archie-core/internal/channels/email"
@@ -14,6 +16,7 @@ import (
"github.com/samcharles93/archie-core/internal/channels/telegram"
"github.com/samcharles93/archie-core/internal/channels/webhook"
"github.com/samcharles93/archie-core/internal/config"
"github.com/samcharles93/archie-core/internal/domain/applystatus"
"github.com/samcharles93/archie-core/internal/domain/health"
"github.com/samcharles93/archie-core/internal/domain/messaging"
"github.com/samcharles93/archie-core/internal/domain/storecontract"
@@ -30,11 +33,44 @@ type deps struct {
Chat messaging.ChatContract
Health *health.Registry
Settings *messaging.SettingsCommand
// SettingsSource, when non-nil, enables the live channel-settings
// reconcile: the service re-reads the stored channel-settings resource on
// the apply-status restamp interval and restarts only the channels whose
// settings changed. Nil leaves the service with no live path, which is the
// honest state for a test or a process with no State Store.
SettingsSource chatSettingsSource
// ApplyReporter records what this process applied. Nil reports nothing.
ApplyReporter *applystatus.Reporter
// AppliedVersion is the channel-settings version already running when the
// service starts, so the first reconcile tick does not re-apply it.
AppliedVersion int64
// ReconcileInterval overrides applystatus.RestampInterval, so a test can
// drive the loop without waiting 30 seconds.
ReconcileInterval time.Duration
}
// channelInstance is one composed channel and the runtime state that lets the
// service restart it alone. channel and cancel are guarded by mu; rebuild is
// immutable. The channel swaps when the stored settings change, which is what a
// restart is: build the replacement, stop the old run context, start the new
// instance. A channel with no in-place reload seam is therefore still
// reloadable.
type channelInstance struct {
name string
channel channels.Channel
rebuild func(ResolvedConfig) (channels.Channel, error)
mu sync.Mutex
channel channels.Channel
cancel context.CancelFunc
supervised bool
restartPending bool
}
// current returns the channel instance currently serving, under the lock.
func (c *channelInstance) current() channels.Channel {
c.mu.Lock()
defer c.mu.Unlock()
return c.channel
}
// Service manages the lifecycle of the extracted Messaging Service and its
@@ -55,11 +91,26 @@ type Service struct {
log *slog.Logger
chat messaging.ChatContract
health *health.Registry
channels []channelInstance
channels []*channelInstance
// settingsSource re-reads the stored channel settings. Nil disables the
// reconcile loop.
settingsSource chatSettingsSource
// reporter records the applied channel-settings version. Nil reports
// nothing.
reporter *applystatus.Reporter
// appliedVersion is the channel-settings version the running channels were
// built from, so a reconcile tick does not re-apply its own work.
appliedVersion atomic.Int64
// reconcileInterval is how often the stored settings are re-read.
reconcileInterval time.Duration
mu sync.Mutex
running bool
stop func()
// wg tracks the per-channel supervisors, so a shutdown waits for every
// channel to stop before Start returns. A restart can start a replacement
// supervisor, which is why it lives here rather than in Start's frame.
wg sync.WaitGroup
}
// compose builds the Service from resolved configuration. ctx is the service
@@ -69,57 +120,122 @@ type Service struct {
// dropped: a front-end the operator configured must never silently not run.
func compose(ctx context.Context, d deps) (*Service, error) {
srv := &Service{
cfg: d.Config,
log: d.Log,
chat: d.Chat,
health: d.Health,
cfg: d.Config,
log: d.Log,
chat: d.Chat,
health: d.Health,
settingsSource: d.SettingsSource,
reporter: d.ApplyReporter,
reconcileInterval: d.ReconcileInterval,
}
srv.appliedVersion.Store(d.AppliedVersion)
if srv.reconcileInterval <= 0 {
srv.reconcileInterval = applystatus.RestampInterval
}
if d.Config.TelegramToken != "" {
if len(d.Config.Telegram.AllowedUserIDs) == 0 {
d.Log.Warn("chat.telegram has no allowed_user_ids: every sender will be rejected. " +
"Add your Telegram user id to chat.telegram.allowed_user_ids to enable the bot.")
}
tg := telegram.New(d.Config.TelegramToken, d.Config.Telegram.AllowedUserIDs, d.Log)
tg.Settings = d.Settings
configureTelegram(ctx, tg, d.Config, d.Chat, d.Log)
if err := srv.add("telegram", tg, telegramValidateConfigMap(d.Config.Telegram)); err != nil {
return nil, err
}
instances, err := composeChannels(ctx, d)
if err != nil {
return nil, err
}
if d.Config.Email.ListenAddr != "" {
em := email.New(d.Config.Email.ListenAddr, d.Config.Email.RelayAddr, d.Log)
if err := srv.add("email", em, map[string]any{
"listen_addr": d.Config.Email.ListenAddr,
"relay_addr": d.Config.Email.RelayAddr,
}); err != nil {
return nil, err
}
}
if d.Config.WebhookAddr != "" {
host, port := parseListenAddr(d.Config.WebhookAddr, "0.0.0.0", 8644)
wh := webhook.New(host, port, webhookRoutes(d.Config.Webhook, d.Config.WebhookSecret), d.Log)
if err := srv.add("webhook", wh, map[string]any{"host": host, "port": port}); err != nil {
return nil, err
}
}
srv.channels = instances
srv.status = status.NewManager(channelDescriptors(srv.channels))
srv.statusWriter = d.ChannelStatus
srv.publish = make(chan struct{}, 1)
return srv, nil
}
// add validates a channel against its own ConfigSchema contract and
// registers it.
func (s *Service) add(name string, ch channels.Channel, cfg map[string]any) error {
if err := ch.ValidateConfig(cfg); err != nil {
return fmt.Errorf("chat.%s config invalid: %w", name, err)
// composeChannels builds the configured channel set. Each instance carries the
// factory a restart uses to build a replacement from a new resolved
// configuration, which is what makes a channel with no in-place reload seam
// still restarted alone rather than through a process restart.
func composeChannels(ctx context.Context, d deps) ([]*channelInstance, error) {
instances := make([]*channelInstance, 0, 3)
if d.Config.TelegramToken != "" {
instance, err := composeTelegram(ctx, d)
if err != nil {
return nil, err
}
instances = append(instances, instance)
}
s.channels = append(s.channels, channelInstance{name: name, channel: ch})
return nil
if d.Config.Email.ListenAddr != "" {
instance, err := composeEmail(d)
if err != nil {
return nil, err
}
instances = append(instances, instance)
}
if d.Config.WebhookAddr != "" {
instance, err := composeWebhook(d)
if err != nil {
return nil, err
}
instances = append(instances, instance)
}
return instances, nil
}
func composeTelegram(ctx context.Context, d deps) (*channelInstance, error) {
if len(d.Config.Telegram.AllowedUserIDs) == 0 {
d.Log.Warn("chat.telegram has no allowed_user_ids: every sender will be rejected. " +
"Add your Telegram user id to chat.telegram.allowed_user_ids to enable the bot.")
}
build := func(cfg ResolvedConfig) (channels.Channel, error) {
if cfg.TelegramToken == "" {
return nil, nil
}
tg := telegram.New(cfg.TelegramToken, cfg.Telegram.AllowedUserIDs, d.Log)
tg.Settings = d.Settings
configureTelegram(ctx, tg, cfg, d.Chat, d.Log)
if err := tg.ValidateConfig(telegramValidateConfigMap(cfg.Telegram)); err != nil {
return nil, fmt.Errorf("chat.telegram config invalid: %w", err)
}
return tg, nil
}
ch, err := build(d.Config)
if err != nil {
return nil, err
}
return &channelInstance{name: "telegram", channel: ch, rebuild: build}, nil
}
func composeEmail(d deps) (*channelInstance, error) {
build := func(cfg ResolvedConfig) (channels.Channel, error) {
if cfg.Email.ListenAddr == "" {
return nil, nil
}
em := email.New(cfg.Email.ListenAddr, cfg.Email.RelayAddr, d.Log)
if err := em.ValidateConfig(map[string]any{
"listen_addr": cfg.Email.ListenAddr,
"relay_addr": cfg.Email.RelayAddr,
}); err != nil {
return nil, fmt.Errorf("chat.email config invalid: %w", err)
}
return em, nil
}
ch, err := build(d.Config)
if err != nil {
return nil, err
}
return &channelInstance{name: "email", channel: ch, rebuild: build}, nil
}
func composeWebhook(d deps) (*channelInstance, error) {
build := func(cfg ResolvedConfig) (channels.Channel, error) {
if cfg.WebhookAddr == "" {
return nil, nil
}
host, port := parseListenAddr(cfg.WebhookAddr, "0.0.0.0", 8644)
wh := webhook.New(host, port, webhookRoutes(cfg.Webhook, cfg.WebhookSecret), d.Log)
if err := wh.ValidateConfig(map[string]any{"host": host, "port": port}); err != nil {
return nil, fmt.Errorf("chat.webhook config invalid: %w", err)
}
return wh, nil
}
ch, err := build(d.Config)
if err != nil {
return nil, err
}
return &channelInstance{name: "webhook", channel: ch, rebuild: build}, nil
}
// telegramValidateConfigMap builds the map Gateway.ValidateConfig expects
@@ -180,24 +296,17 @@ func (s *Service) Start(ctx context.Context) error {
// opened during startup shows configured channels rather than none.
s.signalPublish()
}
if s.settingsSource != nil {
go s.reconcileLoop(ctx)
}
var wg sync.WaitGroup
for _, ch := range s.channels {
c := ch
wg.Go(func() {
s.log.Info("starting channel", "name", c.name)
if err := c.channel.Start(ctx, s.chat, s.lifecycleFor(c.name)); err != nil && ctx.Err() == nil {
// The channel could not run: say so on the operator's surface, not
// only in the log, which is the difference between a dashboard that
// shows a dead channel and one that shows nothing at all.
s.status.MarkFailed(c.name, err.Error())
s.log.Error("channel stopped with error", "name", c.name, "err", err)
return
}
// A channel that returns without an error has stopped, either because
// the service was asked to or because its own loop ended.
s.status.MarkStopped(c.name, "")
})
c.mu.Lock()
c.supervised = true
c.mu.Unlock()
s.wg.Add(1)
go s.runChannel(ctx, c)
}
// Every configured channel has been handed its lifetime and the service
@@ -214,18 +323,10 @@ func (s *Service) Start(ctx context.Context) error {
<-ctx.Done()
for _, ch := range s.channels {
shutdownCtx, shutdownCancel := context.WithTimeout(context.WithoutCancel(ctx), s.cfg.Options.ShutdownTimeout)
if err := ch.channel.Stop(shutdownCtx); err != nil {
s.log.Warn("channel stop failed", "name", ch.name, "err", err)
}
shutdownCancel()
}
wg.Wait()
s.wg.Wait()
if s.statusWriter != nil {
// The final report carries the stopped states the loop above recorded.
// The final report carries the stopped states each supervisor recorded.
// Its own context, because ctx is already cancelled.
finalCtx, cancel := context.WithTimeout(context.WithoutCancel(ctx), s.cfg.Options.ShutdownTimeout)
defer cancel()
+216
View File
@@ -0,0 +1,216 @@
package archiemessaging
import (
"context"
"errors"
"fmt"
"slices"
"strings"
"time"
"github.com/samcharles93/archie-core/internal/channels/telegram"
"github.com/samcharles93/archie-core/internal/config"
"github.com/samcharles93/archie-core/internal/infrastructure/controlplanerpc"
"github.com/samcharles93/archie-core/internal/secret"
)
// chatSettingsSource reads the stored channel-settings resource and layers it
// over the file document. *controlplanerpc.Client satisfies it.
type chatSettingsSource interface {
RuntimeChatConfig(ctx context.Context, base config.ChatConfig) (config.ChatConfig, int64, error)
}
// reconcileLoop re-reads the stored channel settings on the apply-status
// restamp interval until ctx ends. A poll rather than a stream: the restamp
// interval is the cadence every process already reports at, the messaging
// process has no other control-plane watch to share a reconnect ladder with,
// and a stored document that cannot be read must leave the running channels
// alone (docs/prds/plugin-settings-live.md).
func (s *Service) reconcileLoop(ctx context.Context) {
ticker := time.NewTicker(s.reconcileInterval)
defer ticker.Stop()
for {
select {
case <-ctx.Done():
return
case <-ticker.C:
if err := s.reconcileOnce(ctx); err != nil {
s.log.Warn("channel settings reconcile failed", "err", err)
}
}
}
}
// reconcileOnce re-reads the stored channel settings and, when the version
// moved, applies the change to the running channels and records the outcome in
// apply status.
//
// The file document is re-resolved first, exactly as the /restart path does, so
// the layering is the same projection boot used: the store overrides the file,
// and a field the store omits keeps the file's value.
func (s *Service) reconcileOnce(ctx context.Context) error {
base, err := Resolve(s.currentConfig().Options, s.log)
if err != nil {
return err
}
layered, version, err := s.settingsSource.RuntimeChatConfig(ctx, config.ChatConfig{
Telegram: base.Telegram, Email: base.Email, Webhook: base.Webhook,
WebhookAddr: base.WebhookAddr, ShowToolCalls: base.ShowToolCalls,
})
if err != nil {
return err
}
if version == 0 || version <= s.appliedVersion.Load() {
return nil
}
next, err := resolveChatSecrets(base, layered)
if err != nil {
// A credential this process cannot resolve is a document rejection: the
// running channels keep serving and the version stays reported as not
// applied, so the operator sees the refusal rather than a false success.
s.report(ctx, version, err)
return err
}
applied, outstanding := s.applyChannelSettings(next)
if applied {
s.appliedVersion.Store(version)
}
s.report(ctx, version, outstanding)
return nil
}
// resolveChatSecrets carries the layered chat settings onto the resolved
// projection, resolving the store's secret references. A channel whose token
// cannot be resolved stays off rather than composing with an empty credential
// (the startup rule, config.go resolveTelegramToken).
func resolveChatSecrets(base ResolvedConfig, layered config.ChatConfig) (ResolvedConfig, error) {
next := base
next.Telegram, next.Email, next.Webhook = layered.Telegram, layered.Email, layered.Webhook
next.WebhookAddr, next.ShowToolCalls = layered.WebhookAddr, layered.ShowToolCalls
secrets := secret.NewRegistry()
token, err := resolveTelegramToken(layered.Telegram, secrets)
if err != nil {
return next, fmt.Errorf("resolve database telegram token: %w", err)
}
next.TelegramToken = token
webhookSecret, err := resolveWebhookSecret(layered.Webhook, secrets)
if err != nil {
return next, fmt.Errorf("resolve database webhook secret: %w", err)
}
next.WebhookSecret = webhookSecret
return next, nil
}
// applyChannelSettings applies a freshly resolved chat config to the running
// channels. A channel restarts only when its own transport settings changed;
// every other channel keeps serving. It reports whether every requested restart
// was carried out, so a rebuild that failed is retried rather than recorded as
// applied.
//
// Machine-level settings do not restart a channel. The chat session defaults
// (workspace, filesystem access, max steps, rate limit) are not part of the
// messaging projection at all, and show_tool_calls is handed to the running
// Telegram gateway in place, so both apply to turns that start after the change
// rather than interrupting one in flight.
//
// The composed set and the listen addresses are fixed for the life of the
// process. Enabling or disabling a channel and rebinding a listen address are
// process bindings; the change is refused and reported for a restart instead of
// applied (docs/prds/runtime-control-plane.md, "Some changes require a
// restart").
func (s *Service) applyChannelSettings(next ResolvedConfig) (bool, error) {
current := s.currentConfig()
effective, outstanding := pinProcessBindings(current, next)
if next.ShowToolCalls != current.ShowToolCalls {
s.applyShowToolCalls(next.ShowToolCalls)
}
applied := true
for _, id := range channelChanges(current, effective) {
if err := s.restartChannel(id, effective); err != nil {
outstanding = append(outstanding, err.Error())
applied = false
}
}
// A failed restart leaves the running configuration in place, so the next
// tick sees the same change and retries it. A change that was carried out
// (including a binding change that was refused and reported) is recorded, so
// it is not applied again on every tick.
if applied {
s.setConfig(effective)
}
if len(outstanding) == 0 {
return applied, nil
}
return applied, errors.New(strings.Join(outstanding, "; "))
}
// applyShowToolCalls updates the running Telegram gateways in place. Nothing
// else consumes the setting, and it is read per reply, so it needs no restart.
func (s *Service) applyShowToolCalls(show bool) {
for _, c := range s.channels {
if g, ok := c.current().(*telegram.Gateway); ok {
g.SetShowToolCalls(show)
}
}
}
// pinProcessBindings holds the values a running process cannot change in place:
// the listen addresses it already bound, and the set of channels it composed.
// The returned configuration keeps the running value for each, and the returned
// problems name what the operator changed and when it takes effect.
func pinProcessBindings(current, next ResolvedConfig) (ResolvedConfig, []string) {
effective := next
var problems []string
if (current.TelegramToken != "") != (next.TelegramToken != "") {
problems = append(problems, "chat.telegram enabled/disabled; takes effect on restart")
effective.Telegram, effective.TelegramToken = current.Telegram, current.TelegramToken
}
if (current.Email.ListenAddr != "") != (next.Email.ListenAddr != "") {
problems = append(problems, "chat.email enabled/disabled; takes effect on restart")
effective.Email = current.Email
}
if (current.WebhookAddr != "") != (next.WebhookAddr != "") {
problems = append(problems, "chat.webhook enabled/disabled; takes effect on restart")
effective.Webhook, effective.WebhookAddr, effective.WebhookSecret = current.Webhook, current.WebhookAddr, current.WebhookSecret
}
if current.Email.ListenAddr != "" && next.Email.ListenAddr != current.Email.ListenAddr {
problems = append(problems, "chat.email.listen_addr changed; takes effect on restart")
effective.Email.ListenAddr = current.Email.ListenAddr
}
if current.WebhookAddr != "" && next.WebhookAddr != current.WebhookAddr {
problems = append(problems, "chat.webhook_addr changed; takes effect on restart")
effective.WebhookAddr = current.WebhookAddr
}
return effective, problems
}
// channelChanges returns the composed channels whose transport settings differ
// between the running and the effective configuration. A channel absent from
// current (not composed) never appears, and an unchanged channel is not
// restarted.
func channelChanges(current, next ResolvedConfig) []string {
var changed []string
if current.TelegramToken != "" &&
(current.TelegramToken != next.TelegramToken || !slices.Equal(current.Telegram.AllowedUserIDs, next.Telegram.AllowedUserIDs)) {
changed = append(changed, "telegram")
}
if current.Email.ListenAddr != "" && current.Email.RelayAddr != next.Email.RelayAddr {
changed = append(changed, "email")
}
if current.WebhookAddr != "" && (current.Webhook != next.Webhook || current.WebhookSecret != next.WebhookSecret) {
changed = append(changed, "webhook")
}
return changed
}
func (s *Service) report(ctx context.Context, version int64, applyErr error) {
if s.reporter == nil {
return
}
s.reporter.Report(ctx, controlplanerpc.ChannelSettingsKind, version, applyErr)
}
@@ -0,0 +1,253 @@
package archiemessaging
import (
"context"
"fmt"
"log/slog"
"os"
"path/filepath"
"strings"
"sync"
"testing"
"time"
"github.com/samcharles93/archie-core/internal/channels/status"
"github.com/samcharles93/archie-core/internal/config"
"github.com/samcharles93/archie-core/internal/domain/applystatus"
"github.com/samcharles93/archie-core/internal/domain/messaging"
"github.com/samcharles93/archie-core/internal/domain/storecontract"
)
// channelByID returns the live instance the service currently serves for id.
func channelByID(t *testing.T, srv *Service, id string) *channelInstance {
t.Helper()
for _, instance := range srv.channels {
if instance.name == id {
return instance
}
}
t.Fatalf("no channel %q composed", id)
return nil
}
// channelAddrs reserves two loopback ports, one per channel.
func channelAddrs(t *testing.T) (emailAddr, webhookAddr string) {
t.Helper()
return fmt.Sprintf("127.0.0.1:%d", freePort(t)), fmt.Sprintf("127.0.0.1:%d", freePort(t))
}
// writeChannelConfig writes a file document with email and webhook enabled. A
// file rather than a hand-built ResolvedConfig, because the reconcile path
// re-resolves the file exactly as production does; a hand-built config would
// prove nothing about that half.
func writeChannelConfig(t *testing.T, emailAddr, webhookAddr, route string) string {
t.Helper()
path := filepath.Join(t.TempDir(), "config.toml")
content := fmt.Sprintf("bot_user = \"testbot\"\n\n[chat]\nwebhook_addr = %q\n\n[chat.email]\nlisten_addr = %q\n\n[chat.webhook]\npath = %q\n",
webhookAddr, emailAddr, route)
if err := os.WriteFile(path, []byte(content), 0o600); err != nil {
t.Fatal(err)
}
return path
}
func resolveChannels(t *testing.T, emailAddr, webhookAddr, route string) ResolvedConfig {
t.Helper()
cfg, err := Resolve(
Options{Config: writeChannelConfig(t, emailAddr, webhookAddr, route), ShutdownTimeout: 2 * time.Second},
slog.New(slog.DiscardHandler),
)
if err != nil {
t.Fatalf("Resolve: %v", err)
}
return cfg
}
// startComposed starts a composed service and waits for its channels to run.
func startComposed(t *testing.T, d deps) *Service {
t.Helper()
srv, err := compose(t.Context(), d)
if err != nil {
t.Fatalf("compose: %v", err)
}
ctx, cancel := context.WithCancel(t.Context())
started := make(chan error, 1)
go func() { started <- srv.Start(ctx) }()
t.Cleanup(func() {
srv.Stop()
cancel()
select {
case err := <-started:
if err != nil {
t.Errorf("srv.Start returned %v, want a clean stop", err)
}
case <-time.After(3 * time.Second):
t.Error("srv.Start did not return after Stop")
}
})
waitForState(t, srv, "email", status.StateRunning)
waitForState(t, srv, "webhook", status.StateRunning)
return srv
}
// startChannels composes and starts email and webhook. Those two are the
// channels a test can run without an external service, and a restart of one is
// observable as the channel pointer changing.
func startChannels(t *testing.T) *Service {
t.Helper()
emailAddr, webhookAddr := channelAddrs(t)
return startComposed(t, deps{
Config: resolveChannels(t, emailAddr, webhookAddr, "/a"),
Log: slog.New(slog.DiscardHandler),
Chat: &recordingChatContract{routes: make(chan messaging.Inbound, 1), reply: "ok"},
})
}
// TestApplyChannelSettingsRestartsOnlyTheChangedChannel is the bead's core
// property: a change to one channel restarts that channel and leaves the others
// serving. Before the per-channel supervisor this could only be a process
// restart, which restarted every channel (archie-core-zfb0.4).
func TestApplyChannelSettingsRestartsOnlyTheChangedChannel(t *testing.T) {
srv := startChannels(t)
webhookBefore := channelByID(t, srv, "webhook").current()
emailBefore := channelByID(t, srv, "email").current()
next := srv.currentConfig()
next.Webhook = config.WebhookRoute{Path: "/b"}
if _, err := srv.applyChannelSettings(next); err != nil {
t.Fatalf("applyChannelSettings: %v", err)
}
if got := channelByID(t, srv, "webhook").current(); got == webhookBefore {
t.Error("webhook did not restart after its route changed")
}
if got := channelByID(t, srv, "email").current(); got != emailBefore {
t.Error("email restarted when only the webhook route changed")
}
}
// TestApplyChannelSettingsRefusesAListenAddressChange is the path where the
// change must NOT apply: a bound listen address is a process binding, and
// rebinding it in place is exactly the restart the epic keeps. The running
// channel is left alone and the change is reported instead.
func TestApplyChannelSettingsRefusesAListenAddressChange(t *testing.T) {
srv := startChannels(t)
before := channelByID(t, srv, "webhook").current()
next := srv.currentConfig()
next.WebhookAddr = fmt.Sprintf("127.0.0.1:%d", freePort(t))
_, err := srv.applyChannelSettings(next)
if err == nil {
t.Fatal("applyChannelSettings error = nil, want the listen-address change reported")
}
if !strings.Contains(err.Error(), "webhook_addr") {
t.Errorf("error = %v, want it to name chat.webhook_addr", err)
}
if got := channelByID(t, srv, "webhook").current(); got != before {
t.Error("webhook restarted for a listen-address change; a bound address must not move in place")
}
}
// TestApplyChannelSettingsRestartsNothingWhenNothingChanged: the other half of
// "only the channels whose settings changed" -- a tick that carries no change
// must not restart a healthy channel.
func TestApplyChannelSettingsRestartsNothingWhenNothingChanged(t *testing.T) {
srv := startChannels(t)
webhookBefore := channelByID(t, srv, "webhook").current()
emailBefore := channelByID(t, srv, "email").current()
if _, err := srv.applyChannelSettings(srv.currentConfig()); err != nil {
t.Fatalf("applyChannelSettings: %v", err)
}
if channelByID(t, srv, "webhook").current() != webhookBefore {
t.Error("webhook restarted with no settings change")
}
if channelByID(t, srv, "email").current() != emailBefore {
t.Error("email restarted with no settings change")
}
}
// fakeChannelSettings is the stored channel-settings resource a test owns. It
// returns the document a real store would: the file's values with the stored
// route layered over them, so the reconcile's layering is exercised rather than
// bypassed.
type fakeChannelSettings struct {
mu sync.Mutex
settings config.ChatConfig
version int64
}
func (f *fakeChannelSettings) RuntimeChatConfig(_ context.Context, base config.ChatConfig) (config.ChatConfig, int64, error) {
f.mu.Lock()
defer f.mu.Unlock()
out := base
out.Webhook = f.settings.Webhook
return out, f.version, nil
}
func (f *fakeChannelSettings) set(version int64, route config.WebhookRoute) {
f.mu.Lock()
defer f.mu.Unlock()
f.version, f.settings.Webhook = version, route
}
// recordingApplyStore records the apply-status writes the service makes.
type recordingApplyStore struct {
mu sync.Mutex
writes []storecontract.ApplyStatus
}
func (s *recordingApplyStore) PutApplyStatus(_ context.Context, st storecontract.ApplyStatus) error {
s.mu.Lock()
defer s.mu.Unlock()
s.writes = append(s.writes, st)
return nil
}
func (s *recordingApplyStore) ListApplyStatus(context.Context) ([]storecontract.ApplyStatus, error) {
return nil, nil
}
func (s *recordingApplyStore) appliedVersion() int64 {
s.mu.Lock()
defer s.mu.Unlock()
var version int64
for _, st := range s.writes {
version = max(version, st.AppliedVersion)
}
return version
}
// TestReconcileAppliesANewStoredVersion drives the whole live path: the service
// reads the stored channel settings on its interval, sees a newer version, and
// restarts the channel it changed. The applied version reaches apply status, so
// the dashboard stops showing the change as pending.
func TestReconcileAppliesANewStoredVersion(t *testing.T) {
emailAddr, webhookAddr := channelAddrs(t)
source := &fakeChannelSettings{}
source.set(1, config.WebhookRoute{Path: "/a"})
store := &recordingApplyStore{}
srv := startComposed(t, deps{
Config: resolveChannels(t, emailAddr, webhookAddr, "/a"),
Log: slog.New(slog.DiscardHandler),
Chat: &recordingChatContract{routes: make(chan messaging.Inbound, 1), reply: "ok"},
SettingsSource: source,
ApplyReporter: applystatus.New(applystatus.Messaging, store, slog.New(slog.DiscardHandler)),
AppliedVersion: 1,
ReconcileInterval: 5 * time.Millisecond,
})
before := channelByID(t, srv, "webhook").current()
source.set(2, config.WebhookRoute{Path: "/b"})
deadline := time.Now().Add(3 * time.Second)
for time.Now().Before(deadline) {
if channelByID(t, srv, "webhook").current() != before && store.appliedVersion() >= 2 {
return
}
time.Sleep(5 * time.Millisecond)
}
t.Fatalf("after the stored version moved to 2: channel replaced=%v, applied version=%d; want both",
channelByID(t, srv, "webhook").current() != before, store.appliedVersion())
}
+19 -7
View File
@@ -36,6 +36,13 @@ func Run(ctx context.Context, o Options) error {
// channelStatusStore is the same client, held where compose can reach it: the
// channel report is published to the store this process already dials.
var channelStatusStore storecontract.ChannelStatusStore
// settingsSource and applyReporter are the live channel-settings path: the
// control-plane client to re-read the resource, and the record of what this
// process applied. Both stay nil without a State Store, which leaves the
// service with no live path.
var settingsSource chatSettingsSource
var applyReporter *applystatus.Reporter
var appliedVersion int64
if cfg.Options.StateStore.Target != "" {
stateStore, closeClient, dialErr := staterpc.Dial(cfg.Options.StateStore.Target, cfg.Options.StateStore.Token)
if dialErr != nil {
@@ -43,7 +50,8 @@ func Run(ctx context.Context, o Options) error {
}
closeStateStore = closeClient
defer closeStateStore()
chatSettings, channelVersion, settingsErr := controlplanerpc.NewRPCClient(stateStore.ControlPlane()).RuntimeChatConfig(ctx, config.ChatConfig{
controlPlaneClient := controlplanerpc.NewRPCClient(stateStore.ControlPlane())
chatSettings, channelVersion, settingsErr := controlPlaneClient.RuntimeChatConfig(ctx, config.ChatConfig{
Telegram: cfg.Telegram, Email: cfg.Email, Webhook: cfg.Webhook, WebhookAddr: cfg.WebhookAddr, ShowToolCalls: cfg.ShowToolCalls,
})
if settingsErr != nil {
@@ -64,15 +72,19 @@ func Run(ctx context.Context, o Options) error {
}
settings = messaging.NewSettingsCommand(messagingControlPlane{client: stateStore.ControlPlane()}).WithIdentities(stateStore)
channelStatusStore = stateStore
settingsSource, applyReporter, appliedVersion = controlPlaneClient, reporter, channelVersion
}
srv, err := compose(ctx, deps{
ChannelStatus: channelStatusStore,
Config: cfg,
Log: log,
Chat: chat,
Health: health,
Settings: settings,
ChannelStatus: channelStatusStore,
Config: cfg,
Log: log,
Chat: chat,
Health: health,
Settings: settings,
SettingsSource: settingsSource,
ApplyReporter: applyReporter,
AppliedVersion: appliedVersion,
})
if err != nil {
return err
+150
View File
@@ -0,0 +1,150 @@
package archiemessaging
import (
"context"
"errors"
"fmt"
"github.com/samcharles93/archie-core/internal/channels"
)
// runChannel supervises one channel for the life of the service. It starts the
// channel and, when the channel stops because a restart was requested, starts
// the replacement. A channel that returns on its own is recorded and not
// restarted, so a channel whose configuration cannot run does not spin.
//
// The supervisor is per channel rather than one loop over every channel
// because a channel-settings change must restart exactly the channel whose
// settings changed: cancelling one run context stops one channel and leaves the
// rest serving (archie-core-zfb0.4).
func (s *Service) runChannel(ctx context.Context, c *channelInstance) {
defer s.wg.Done()
for {
c.mu.Lock()
ch := c.channel
runCtx, cancel := context.WithCancel(ctx)
c.cancel = cancel
c.mu.Unlock()
s.log.Info("starting channel", "name", c.name)
err := ch.Start(runCtx, s.chat, s.lifecycleFor(c.name))
// Decide the next state under the same lock a restart takes, so a
// restart racing this exit is either seen here (restartPending) or
// observed as a channel that is no longer supervised.
c.mu.Lock()
c.cancel = nil
shutdown := ctx.Err() != nil
restart := c.restartPending
if !shutdown && restart {
c.restartPending = false
}
if shutdown || !restart {
c.supervised = false
}
c.mu.Unlock()
cancel()
// Stop is safe to call after the run context was cancelled: each
// channel's Start returns having closed what it opened, and Stop is the
// idempotent tail for the paths that leave a listener behind.
s.stopChannel(ctx, c.name, ch)
if shutdown {
if err != nil && !errors.Is(err, context.Canceled) {
s.status.MarkFailed(c.name, err.Error())
} else {
s.status.MarkStopped(c.name, "")
}
return
}
if restart {
// The replacement is already in c.channel (see restartChannel); the
// next pass starts it.
continue
}
// The channel stopped on its own.
if err != nil {
s.status.MarkFailed(c.name, err.Error())
s.log.Error("channel stopped with error", "name", c.name, "err", err)
} else {
s.status.MarkStopped(c.name, "")
}
return
}
}
// restartChannel stops the running channel id and starts a replacement built
// from next. The replacement is built before the running channel is touched, so
// a configuration this build cannot run leaves the running channel serving
// (the validate-before-switch rule in docs/prds/runtime-control-plane.md).
//
// A channel that is not currently supervised is not restarted: it already
// stopped on its own, and resurrecting it here would race that exit. The caller
// reports it.
func (s *Service) restartChannel(id string, next ResolvedConfig) error {
c := s.instanceByID(id)
if c == nil {
return fmt.Errorf("no channel %q", id)
}
if c.rebuild == nil {
return fmt.Errorf("channel %q cannot be rebuilt", id)
}
replacement, err := c.rebuild(next)
if err != nil {
return fmt.Errorf("rebuild %q: %w", id, err)
}
if replacement == nil {
return fmt.Errorf("channel %q has no configuration", id)
}
c.mu.Lock()
if !c.supervised {
c.mu.Unlock()
return fmt.Errorf("channel %q is not running", id)
}
c.channel = replacement
c.restartPending = true
cancel := c.cancel
c.mu.Unlock()
if cancel != nil {
cancel()
}
return nil
}
// stopChannel closes one channel's listener with its own bounded context, so a
// stop that hangs cannot hold the shutdown past the configured timeout. The
// parent context is stripped of its cancellation because this runs on the
// shutdown path, where the parent is already cancelled.
func (s *Service) stopChannel(ctx context.Context, name string, ch channels.Channel) {
timeout := s.currentConfig().Options.ShutdownTimeout
if timeout <= 0 {
timeout = defaultShutdownTimeout
}
stopCtx, cancel := context.WithTimeout(context.WithoutCancel(ctx), timeout)
defer cancel()
if err := ch.Stop(stopCtx); err != nil {
s.log.Warn("channel stop failed", "name", name, "err", err)
}
}
func (s *Service) instanceByID(id string) *channelInstance {
for _, c := range s.channels {
if c.name == id {
return c
}
}
return nil
}
func (s *Service) currentConfig() ResolvedConfig {
s.mu.Lock()
defer s.mu.Unlock()
return s.cfg
}
func (s *Service) setConfig(cfg ResolvedConfig) {
s.mu.Lock()
s.cfg = cfg
s.mu.Unlock()
}
@@ -8,6 +8,7 @@ import (
"path/filepath"
"time"
"github.com/samcharles93/archie-core/internal/buildinfo"
"github.com/samcharles93/archie-core/internal/channels/telegram"
"github.com/samcharles93/archie-core/internal/domain/messaging"
"github.com/samcharles93/archie-core/internal/installtype"
@@ -17,13 +18,15 @@ import (
// configureTelegram wires the operator seams telegram.Gateway leaves to its
// composition root. Which process may satisfy each one is settled in
// docs/architecture/migration-decisions.md ("Telegram operator surface after
// extraction"); notably RunningVersions, ReleaseAnnouncements and Dangerous
// stay nil deliberately, because nothing this process can observe would make
// them true.
// extraction"); notably ReleaseAnnouncements and Dangerous stay nil, because
// nothing this process can observe would make them true. RunningVersions is
// supplied here from this process's own build stamp; see
// messagingRunningVersions.
func configureTelegram(ctx context.Context, g *telegram.Gateway, cfg ResolvedConfig, chat messaging.ChatContract, log *slog.Logger) {
g.Version = gatewayVersionReporter(ctx, chat, cfg.Options.DependencyTimeout)
g.SetShowToolCalls(cfg.ShowToolCalls)
g.Reload = telegramReloader(cfg.Options, log)
g.RunningVersions = messagingRunningVersions
if updates := updateService(cfg); updates != nil {
g.Updates = updates
@@ -33,6 +36,28 @@ func configureTelegram(ctx context.Context, g *telegram.Gateway, cfg ResolvedCon
}
}
// messagingRunningVersions reports the component versions this process can
// vouch for, for checking a relayed update report against
// (releaseupdate.Report.Verify).
//
// archie-messaging is one of the processes in the archied release set (the
// installer's GATEWAY_SERVICES), built from the same archied/v* tag and stamped
// with buildinfo.Version by every build site (install.sh, the updater, the
// release workflow and Taskfile). Its own compiled-in build is therefore
// evidence that the release the installer claims actually reached this host: a
// binary the update did not replace still reports the version it was built
// from, which Verify reads as drift rather than false success. Reading the
// version from the Gateway instead would report a different process's stamp,
// which is the failure mode the check exists to catch (see
// docs/architecture/migration-decisions.md).
//
// The agent is deliberately absent: every archie-agent process is task-scoped
// and observed only through daemon.AgentStatus, which this process does not
// hold, so it must report Unverified rather than be guessed.
func messagingRunningVersions() map[string]string {
return map[string]string{releaseupdate.ComponentDaemon: buildinfo.Version}
}
// gatewayVersionReporter renders /version from the Gateway's own build block.
// It is read per call rather than captured at compose so an upgraded Gateway
// is reported without restarting this process, and bounded by the same
@@ -8,9 +8,11 @@ import (
"strings"
"testing"
"github.com/samcharles93/archie-core/internal/buildinfo"
"github.com/samcharles93/archie-core/internal/channels/telegram"
"github.com/samcharles93/archie-core/internal/config"
"github.com/samcharles93/archie-core/internal/domain/messaging"
"github.com/samcharles93/archie-core/internal/releaseupdate"
)
type versionChatContract struct {
@@ -84,17 +86,32 @@ func TestTelegramVersionReportsAnUnreachableGateway(t *testing.T) {
}
}
// TestTelegramRunningVersionsStaysUnwired pins the decision recorded in
// docs/architecture/migration-decisions.md: RunningVersions turns an
// installer's claim into a checked one, and only a component's own compiled-in
// build can vouch for itself. This process knows neither archied's build nor
// the observed archie-agent version, so reporting anything here would
// manufacture the false success the check exists to catch.
func TestTelegramRunningVersionsStaysUnwired(t *testing.T) {
// TestTelegramRunningVersionsReportsTheMessagingBuild pins that the Messaging
// Service vouches for the daemon component from its own compiled-in release
// stamp. Every host binary in the archied release set -- archie-messaging
// included -- is built from the same archied/v* tag and stamped with
// buildinfo.Version by every build site (install.sh, the updater, the release
// workflow and Taskfile), so this process's own build is evidence that the
// release actually reached the host, and an old binary still running reports
// the old version as drift. Reading it from the Gateway instead would be a
// different process's stamp, which is the false success the check exists to
// catch.
//
// It reports nothing for the agent: every archie-agent process is task-scoped
// and observed only through daemon.AgentStatus, which this process does not
// hold. Leaving it out makes it Unverified rather than guessed.
func TestTelegramRunningVersionsReportsTheMessagingBuild(t *testing.T) {
gateway := telegramConfig(t, ResolvedConfig{}, &versionChatContract{})
if gateway.RunningVersions != nil {
t.Fatal("telegram RunningVersions is wired; update reports must stay unverified claims")
if gateway.RunningVersions == nil {
t.Fatal("telegram RunningVersions = nil, want the Messaging Service's own release stamp")
}
running := gateway.RunningVersions()
if got := running[releaseupdate.ComponentDaemon]; got != buildinfo.Version {
t.Fatalf("running[%q] = %q, want buildinfo.Version = %q", releaseupdate.ComponentDaemon, got, buildinfo.Version)
}
if _, ok := running[releaseupdate.ComponentAgent]; ok {
t.Fatalf("running vouches for %q, which this process cannot observe: %#v", releaseupdate.ComponentAgent, running)
}
if gateway.Dangerous != nil {
t.Fatal("telegram Dangerous is wired; no composition has ever supplied a command authority")
+9
View File
@@ -113,6 +113,15 @@ func compose(d deps) *webui.Server {
if bindings, ok := d.Store.(storecontract.BindingStore); ok {
srv.Bindings = bindings
}
// Harness OAuth token sets are a ratified State Store contract, and the
// same client already carries them. Withholding it would leave the
// harness page answering 503 on a deployment that has the store
// (archie-core-egkf.7). The setup terminal itself is not wired here: its
// implementation owns a container, which this process must not link, so
// the route stays a documented 503 until a daemon-side contract exists.
if secrets, ok := d.Store.(storecontract.HarnessSecretStore); ok {
srv.HarnessSecrets = secrets
}
if eventTypes, ok := d.Store.(storecontract.EventTypeStore); ok {
srv.EventTypes = eventTypes
}
@@ -105,6 +105,19 @@ func TestRoutesWithoutAContractDegradeExplicitly(t *testing.T) {
name: "capture inspector", method: http.MethodGet, path: "/api/captures",
want: http.StatusOK, emptyJSON: `"enabled":true`,
},
{
// The harness page's two routes: the bindings read needs the
// control-plane resource and the setup terminal needs a transport
// this process cannot own. In this composition neither is wired,
// so both degrade explicitly rather than rendering an empty page
// (archie-core-egkf.7).
name: "harness bindings", method: http.MethodGet, path: "/api/harness/bindings",
want: http.StatusServiceUnavailable,
},
{
name: "harness setup terminal", method: http.MethodGet, path: "/api/harness/terminal?profile=claude-kit",
want: http.StatusServiceUnavailable,
},
{
// The browser reads this to decide which sections to show at
// all, so it is the one route that must answer here.
+2 -2
View File
@@ -72,11 +72,11 @@ func (f *fakeChat) Cancel(context.Context, string) (gateway.ChatCancellation, er
func (f *fakeChat) SetPersona(context.Context, string, string) (bool, error) { return false, nil }
func (f *fakeChat) ApplyTaskAction(context.Context, string, int64, taskstate.Action) (gateway.TaskActionResult, error) {
func (f *fakeChat) ApplyTaskAction(context.Context, string, int64, taskstate.Action, taskactions.ActionPayload) (gateway.TaskActionResult, error) {
return gateway.TaskActionResult{}, nil
}
func (f *fakeChat) ApplyOperatorTaskAction(context.Context, taskactions.Actor, int64, taskstate.Action) (gateway.TaskActionResult, error) {
func (f *fakeChat) ApplyOperatorTaskAction(context.Context, taskactions.Actor, int64, taskstate.Action, taskactions.ActionPayload) (gateway.TaskActionResult, error) {
return gateway.TaskActionResult{}, nil
}
@@ -197,11 +197,11 @@ func TestRegistryRoutesAndValidatesDefinitions(t *testing.T) {
found := false
for _, descriptor := range catalog.Resources {
if descriptor.Kind == ContainerRuntimePoliciesKind {
found = descriptor.ApplyMode == "restart-required"
found = descriptor.ApplyMode == "live"
}
}
if !found {
t.Fatal("container runtime policies missing restart-required descriptor")
t.Fatal("container runtime policies missing live descriptor")
}
_, err = server.Command(t.Context(), &pb.CommandRequest{Kind: ContainerRuntimePoliciesKind, Command: "replace", ValueJson: []byte(`{"pull_policy":"sometimes"}`), Actor: "test", Source: "test", RequestId: "bad", ExpectedVersion: 1})
@@ -0,0 +1,80 @@
package controlplane
import (
"encoding/json"
"os"
"path/filepath"
"testing"
)
// The dashboard mirrors three of this package's validators by hand
// (ui/src/settings/validation.ts), and the model-role rule had already drifted:
// the dashboard required `^[^/\s]+\/\S+$` while the server only asks for a
// slash, so the dashboard refused values this package accepts. The fixture at
// testdata/dashboard-validators.json is the tie: this test runs the real server
// validators over it, and ui/test/control-plane-validator-parity.test.ts runs
// the real dashboard validators over the same cases. Changing a rule on one
// side alone fails one of the two tests (archie-core-ui-dashboard-3).
type dashboardValidatorFixture struct {
Description string `json:"description"`
Cases map[string][]dashboardValidatorCase `json:"cases"`
}
type dashboardValidatorCase struct {
Why string `json:"why"`
Valid bool `json:"valid"`
Value json.RawMessage `json:"value"`
}
func TestDashboardValidatorsAgreeWithTheSharedFixture(t *testing.T) {
raw, err := os.ReadFile(filepath.Join("testdata", "dashboard-validators.json"))
if err != nil {
t.Fatalf("read the shared validator fixture: %v", err)
}
var fixture dashboardValidatorFixture
if err := json.Unmarshal(raw, &fixture); err != nil {
t.Fatalf("decode the shared validator fixture: %v", err)
}
if fixture.Description == "" {
t.Error("the fixture has no description; it is the contract the dashboard mirror is read against")
}
// Each entry is the real entry point a Definition wires into Decode, keyed
// to match the fixture (and the dashboard's own validator map).
validators := map[string]func([]byte) error{
ModelRoleAssignmentsKind: validateModelRoles,
SchedulingPolicyKind: validateScheduling,
SchedulesKind: func(input []byte) error { return validateAs(input, validateSchedules) },
}
for kind, validate := range validators {
cases, ok := fixture.Cases[kind]
if !ok {
t.Errorf("fixture has no cases for %s, so nothing pins the dashboard mirror of it", kind)
continue
}
var accepted, refused int
for _, testCase := range cases {
err := validate(testCase.Value)
switch {
case testCase.Valid && err != nil:
t.Errorf("%s: fixture case %q is valid, but the server refused it: %v", kind, testCase.Why, err)
case !testCase.Valid && err == nil:
t.Errorf("%s: fixture case %q is invalid, but the server accepted it", kind, testCase.Why)
case testCase.Valid:
accepted++
default:
refused++
}
}
if accepted == 0 || refused == 0 {
t.Errorf("%s: the fixture must pin both an accepted and a refused case (accepted=%d refused=%d)", kind, accepted, refused)
}
}
for kind := range fixture.Cases {
if _, ok := validators[kind]; !ok {
t.Errorf("fixture names %s, which this package has no validator for", kind)
}
}
}
+1 -1
View File
@@ -8,7 +8,7 @@ import (
// Schemas are not declared here: each Definition names its document type, and
// Definition.Descriptor derives the JSON Schema from it (internal/app/controlplane/schema.go).
func builtinDefinitions(steps workflow.StepRegistry) []Definition {
definitions := []Definition{workflowDefinition(), workflowDefinitionsDefinition(steps), workflowEnablementDefinition(), personaDefinition(), scheduleDefinition(), agentProfileDefinition(), credentialBindingsDefinition()}
definitions := []Definition{workflowDefinition(), workflowDefinitionsDefinition(steps), workflowEnablementDefinition(), personaDefinition(), scheduleDefinition(), agentProfileDefinition(), credentialBindingsDefinition(), reviewSettingsDefinition()}
definitions = append(definitions, modelDefinitions()...)
definitions = append(definitions, operationalDefinitions()...)
return definitions
@@ -41,15 +41,33 @@ type pluginSettings struct {
func operationalDefinitions() []Definition {
return []Definition{
{Kind: RepositoryPoliciesKind, Title: "Repository policies", ApplyMode: "live", Document: []config.Repo{}, Seed: func(cfg config.Config) any { return cfg.Repos }, Validate: validateRepositories},
{Kind: ChannelSettingsKind, Title: "Channel settings", ApplyMode: "restart-required", Document: channelSettings{}, Seed: seedChannels, Validate: validateChannels, Normalize: normalizeChannels},
{Kind: ChannelSettingsKind, Title: "Channel settings", ApplyMode: "live", Document: channelSettings{}, Seed: seedChannels, Validate: validateChannels, Normalize: normalizeChannels},
{Kind: SchedulingPolicyKind, Title: "Scheduling policy", ApplyMode: "live", Document: schedulingPolicy{}, Seed: func(cfg config.Config) any {
return schedulingPolicy{PollInterval: cfg.PollInterval.Std().String(), MaxRetries: cfg.MaxRetries, Label: &cfg.Label, Dispatch: cfg.Dispatch}
}, Validate: validateScheduling},
{Kind: ToolSettingsKind, Title: "Tool and MCP settings", ApplyMode: "restart-required", Document: toolSettings{}, Seed: seedTools, Validate: validateTools},
{Kind: PluginSettingsKind, Title: "Plugin settings", ApplyMode: "restart-required", Document: pluginSettings{}, Seed: func(cfg config.Config) any {
// ApplyMode is live: the daemon diffs the stored MCP server set against
// the running providers -- connecting added servers, disconnecting
// removed ones and reconnecting changed ones while leaving unchanged
// servers untouched -- and rebuilds the web_fetch and minimax entries
// (archie-core-zfb0.3). A changed server whose new engine fails to start
// is rolled back to the old one, so the refusal is reported rather than
// leaving the server down.
{Kind: ToolSettingsKind, Title: "Tool and MCP settings", ApplyMode: "live", Document: toolSettings{}, Seed: seedTools, Validate: validateTools},
// ApplyMode is live for additions: a stored directory change re-layers live
// and the daemon's reconciliation loads new and changed files without a
// restart. A removal cannot unload Yaegi's interpreter, so it stays an
// outstanding apply-status problem rather than a restart of the whole kind
// (docs/prds/plugin-settings-live.md).
{Kind: PluginSettingsKind, Title: "Plugin settings", ApplyMode: "live", Document: pluginSettings{}, Seed: func(cfg config.Config) any {
return pluginSettings{cfg.PluginDir, cfg.ModuleDir, cfg.SecretEngineDir, cfg.SkillsDir}
}, Validate: validatePluginSettings},
{Kind: ContainerRuntimePoliciesKind, Title: "Container runtime policies", ApplyMode: "restart-required", Document: containerRuntimePolicies{}, Seed: seedContainerPolicies, Validate: validateContainers, Normalize: normalizeContainerPolicies},
// ApplyMode is live: the container pool reads the published config on
// every acquire (image, pull policy, network, max uptime, concurrency
// cap) and the daemon resizes its running dispatcher after the publish,
// so neither a deferred container nor a queued task needs a restart
// (archie-core-zfb0.2). RegistryAuth is file-owned and never part of
// this document, so it stays a boot-resolved value.
{Kind: ContainerRuntimePoliciesKind, Title: "Container runtime policies", ApplyMode: "live", Document: containerRuntimePolicies{}, Seed: seedContainerPolicies, Validate: validateContainers, Normalize: normalizeContainerPolicies},
}
}
@@ -0,0 +1,39 @@
package controlplane
import (
"github.com/samcharles93/archie-core/internal/config"
)
// ReviewSettingsKind is the control-plane resource holding the pr-review
// workflow's two dials: whether a post-worthiness pass runs before findings are
// posted, and whether a human approves the draft before it posts. The file's
// [review] section seeds it; the stored document outranks the file from then on
// (docs/prds/review-settings-resource.md).
const ReviewSettingsKind = "review-settings"
type reviewSettings struct {
PrecisionGate bool `json:"precision_gate" title:"Precision gate" doc:"Run a post-worthiness pass before findings are posted."`
ApproveBeforePost bool `json:"approve_before_post" title:"Approve before posting" doc:"Park a finished review for operator approval before it posts."`
}
func reviewSettingsDefinition() Definition {
return Definition{
Kind: ReviewSettingsKind,
Title: "Review settings",
ApplyMode: "live",
Document: reviewSettings{},
Seed: func(cfg config.Config) any {
return reviewSettings{PrecisionGate: cfg.Review.PrecisionGate, ApproveBeforePost: cfg.Review.ApproveBeforePost}
},
Validate: validateReviewSettings,
}
}
// validateReviewSettings accepts every value, deliberately: both dials are free
// booleans with no cross-field rule, and a stored value reaches its consumers
// through the same layered cfg.Review boot validates. It stays a real function
// rather than an inline no-op so a rule, if one is ever settled, has one
// obvious home (validatePluginSettings is the precedent).
func validateReviewSettings(input []byte) error {
return validateAs(input, func(reviewSettings) error { return nil })
}
@@ -160,6 +160,21 @@ func runtimeConfigFrom(ctx context.Context, reader controlplanerpc.ResourceReade
}); err != nil {
return config.Config{}, nil, err
}
// review-settings owns the pr-review dials once the store carries a value.
// The two fields are assigned individually, not `out.Review = ...`: a field
// added to config.Review later keeps the file's value until the stored
// document carries it (docs/prds/review-settings-resource.md).
if err := layerResource(ctx, reader, versions, ReviewSettingsKind, func(value []byte) error {
var settings reviewSettings
if err := json.Unmarshal(value, &settings); err != nil {
return err
}
out.Review.PrecisionGate = settings.PrecisionGate
out.Review.ApproveBeforePost = settings.ApproveBeforePost
return nil
}); err != nil {
return config.Config{}, nil, err
}
return runtimeToolConfigFrom(ctx, reader, versions, out)
}
@@ -213,9 +228,16 @@ func runtimeToolConfigFrom(ctx context.Context, reader controlplanerpc.ResourceR
// now), and settings() never sets it, so whatever the AgentProfileKind
// layering step below assigns is what's left standing regardless of
// which of the two runs first.
//
// RegistryAuth is the same shape of file-owned field, with no resource of
// its own to restore it later the way CredentialBindingsKind restores
// Credentials: the document cannot carry a secret reference, so it has to
// survive this assignment or the daemon pulls anonymously again.
profiles := out.Containers.Profiles
registryAuth := out.Containers.RegistryAuth
out.Containers = policies.settings()
out.Containers.Profiles = profiles
out.Containers.RegistryAuth = registryAuth
return nil
}); err != nil {
return config.Config{}, nil, err
@@ -537,7 +537,7 @@ func populateEveryField(t *testing.T, value reflect.Value, label string) {
// TestSchedulingPolicySeedCarriesTheLabel: the seed a fresh store writes must
// carry the file document's label, so the pairing a stored trigger requires
// is judgeable at write time from the moment the resource exists.
// TestRuntimeResourceKindsApplyLive pins the four kinds the daemon re-layers
// TestRuntimeResourceKindsApplyLive pins the kinds the daemon re-layers
// live (archie-core-zfb0.1) and the ones a startup-built consumer still
// freezes, so a flip in either direction is a deliberate edit to the
// definition, not a default that drifts.
@@ -553,17 +553,17 @@ func TestRuntimeResourceKindsApplyLive(t *testing.T) {
modes[definition.Kind] = definition.ApplyMode
}
for _, kind := range []string{
ProviderSettingsKind, ModelRoleAssignmentsKind, RepositoryPoliciesKind, SchedulingPolicyKind, AgentProfileKind, CredentialBindingsKind,
ProviderSettingsKind, ModelRoleAssignmentsKind, RepositoryPoliciesKind, SchedulingPolicyKind, AgentProfileKind, CredentialBindingsKind, ToolSettingsKind, PluginSettingsKind, ReviewSettingsKind, ChannelSettingsKind, ContainerRuntimePoliciesKind,
} {
if modes[kind] != "live" {
t.Errorf("%s applies %q, want live: the daemon re-layers this kind on a watch", kind, modes[kind])
}
}
for _, kind := range []string{ToolSettingsKind, PluginSettingsKind, ContainerRuntimePoliciesKind, ChannelSettingsKind} {
if modes[kind] != "restart-required" {
t.Errorf("%s applies %q, want restart-required: a startup-built component still holds it", kind, modes[kind])
}
}
// No control-plane kind is restart-required any more: every kind's consumer
// either re-reads the published config or reconciles on its own process's
// watch. The process bindings that would be restart-required (listen
// addresses, the NATS and database URLs, the state and work directories)
// are not control-plane resource kinds at all.
}
func TestSchedulingPolicySeedCarriesTheLabel(t *testing.T) {
@@ -660,6 +660,33 @@ func TestRuntimeConfigLayersStoredProfiles(t *testing.T) {
}
}
// A stored container-runtime-policies document replaces the file's [containers]
// section wholesale, so a file-owned field that document does not carry -- the
// registry credential resolved at boot -- must be preserved across the
// assignment or it is silently dropped and every private-registry pull is
// anonymous again.
func TestRuntimeConfigPreservesFileOwnedRegistryAuth(t *testing.T) {
base := config.Config{Containers: config.ContainerConfig{
Image: "agent:1",
RegistryAuth: config.SecretRef{Engine: "env", Key: "REGISTRY_AUTH"},
}}
client := NewRPCClient(&runtimeConfigClient{values: map[string]any{
SchedulingPolicyKind: map[string]any{"poll_interval": "2m", "dispatch": map[string]any{"trigger": "assignee"}},
ContainerRuntimePoliciesKind: map[string]any{"image": "agent:2", "pull_policy": "always"},
}})
got, _, err := client.RuntimeConfig(t.Context(), base)
if err != nil {
t.Fatalf("RuntimeConfig: %v", err)
}
if got.Containers.Image != "agent:2" {
t.Fatalf("image = %q, want the stored document's", got.Containers.Image)
}
if want := (config.SecretRef{Engine: "env", Key: "REGISTRY_AUTH"}); got.Containers.RegistryAuth != want {
t.Fatalf("registry_auth = %+v, want the file's %+v preserved across the container-policies layering", got.Containers.RegistryAuth, want)
}
}
// A stored CredentialBindingsKind value replaces the file's bindings
// outright; no stored value at all leaves the file's in effect (the same
// shape TestRuntimeConfigLayersStoredProfiles pins for AgentProfileKind, for
@@ -698,3 +725,85 @@ func TestRuntimeConfigLayersStoredCredentialBindings(t *testing.T) {
t.Fatalf("layering mutated the file document's credentials: %v", got)
}
}
// reviewSettingsKindWire is the kind the review-settings resource is addressed
// by. It is deliberately the literal wire string rather than the package
// constant so this test can state the contract before the constant exists, and
// so a rename of the constant that the layering does not follow fails here
// rather than silently testing a different kind.
const reviewSettingsKindWire = "review-settings"
// The file document's [review] section seeds the review-settings kind, so a
// deployment that never edited the dials runs the values in config.toml
// (docs/prds/review-settings-resource.md, "Seeding").
func TestReviewSettingsSeedCarriesBothDials(t *testing.T) {
definition, ok := testServer(t, nil).definitions[reviewSettingsKindWire]
if !ok {
t.Fatalf("kind %q is not a control-plane resource", reviewSettingsKindWire)
}
seed, err := definition.seededValue(config.Config{Review: config.Review{PrecisionGate: true, ApproveBeforePost: false}})
if err != nil {
t.Fatalf("seededValue: %v", err)
}
var document map[string]any
if err := json.Unmarshal(seed, &document); err != nil {
t.Fatalf("decode seed: %v", err)
}
if document["precision_gate"] != true || document["approve_before_post"] != false {
t.Fatalf("seed = %s, want the file document's precision_gate/approve_before_post", seed)
}
}
// A stored review-settings document outranks the file's [review] from then on,
// and the layered value reaches a run through cfg.Review -- what
// TaskContext.Cfg carries into stagePRPrecisionGate and
// stagePROperatorApproval. The assertion is on cfg.Review after layering, not
// on a struct the test set directly.
func TestRuntimeConfigLayersStoredReviewSettings(t *testing.T) {
base := config.Config{Review: config.Review{PrecisionGate: true, ApproveBeforePost: false}}
t.Run("a stored document outranks the file", func(t *testing.T) {
got, versions, err := runtimeConfigFrom(t.Context(), reviewStoreReader{value: []byte(`{"precision_gate":false,"approve_before_post":true}`)}, base)
if err != nil {
t.Fatalf("runtimeConfigFrom: %v", err)
}
if want := (config.Review{PrecisionGate: false, ApproveBeforePost: true}); got.Review != want {
t.Fatalf("cfg.Review = %+v, want the stored document %+v", got.Review, want)
}
if versions[reviewSettingsKindWire] != 7 {
t.Fatalf("versions[%s] = %d, want the version the store answered with", reviewSettingsKindWire, versions[reviewSettingsKindWire])
}
})
t.Run("no stored value keeps the file's", func(t *testing.T) {
got, _, err := runtimeConfigFrom(t.Context(), absentReader{}, base)
if err != nil {
t.Fatalf("runtimeConfigFrom: %v", err)
}
if want := (config.Review{PrecisionGate: true, ApproveBeforePost: false}); got.Review != want {
t.Fatalf("cfg.Review = %+v, want the file's %+v", got.Review, want)
}
})
if base.Review != (config.Review{PrecisionGate: true, ApproveBeforePost: false}) {
t.Fatalf("layering mutated the file document's review settings: %+v", base.Review)
}
}
// reviewStoreReader answers only the review-settings kind, so this test does
// not have to supply every other kind's document to reach the layering it
// exercises.
type reviewStoreReader struct{ value []byte }
func (r reviewStoreReader) Query(_ context.Context, kind string, decode func([]byte) error) (int64, bool, error) {
if kind != reviewSettingsKindWire {
return 0, false, nil
}
if r.value == nil {
return 0, false, nil
}
if err := decode(r.value); err != nil {
return 0, false, err
}
return 7, true, nil
}
@@ -0,0 +1,169 @@
{
"description": "Shared accept/reject fixtures for the control-plane validators the dashboard hand-mirrors in ui/src/settings/validation.ts. The control plane is the authority: every case's `valid` is what internal/app/controlplane accepts or refuses. The Go test in this package runs the real server validators over these cases; the node test ui/test/control-plane-validator-parity.test.ts runs the real dashboard validators over the same cases. A one-sided change to either rule makes one of the two tests fail, which is the tie that stops the two implementations drifting apart (archie-core-ui-dashboard-3). Add a case here, never only to one side.",
"cases": {
"model-role-assignments": [
{
"why": "a plain provider/model value",
"value": { "builder": "openai/gpt-4o" },
"valid": true
},
{
"why": "a trailing space: the server only asks for a slash, so the dashboard must not refuse it",
"value": { "builder": "openai/gpt-4o " },
"valid": true
},
{
"why": "a leading space: the server only asks for a slash, so the dashboard must not refuse it",
"value": { "builder": " openai/gpt-4o" },
"valid": true
},
{
"why": "a model id with a colon (a real ollama tag), which a stricter shape would wrongly refuse",
"value": { "planner": "ollama/qwen2.5-coder:7b" },
"valid": true
},
{
"why": "no slash at all",
"value": { "builder": "gpt-4o" },
"valid": false
},
{
"why": "an empty model: the server refuses it, so the dashboard must too (SystemModelsPage deletes the key rather than storing one)",
"value": { "builder": "" },
"valid": false
},
{
"why": "a blank role key",
"value": { "": "openai/gpt-4o" },
"valid": false
}
],
"scheduling-policy": [
{
"why": "a positive interval, a valid trigger and the label the trigger needs",
"value": {
"poll_interval": "1m0s",
"max_retries": 0,
"label": "archie",
"dispatch": { "trigger": "assignee" }
},
"valid": true
},
{
"why": "assignee does not require a label",
"value": {
"poll_interval": "30s",
"max_retries": 3,
"dispatch": { "trigger": "assignee" }
},
"valid": true
},
{
"why": "either requires a label, and one is present",
"value": {
"poll_interval": "1h30m",
"label": "archie",
"dispatch": { "trigger": "either" }
},
"valid": true
},
{
"why": "a fractional duration the server's parser accepts",
"value": {
"poll_interval": "1.5h",
"dispatch": { "trigger": "assignee" }
},
"valid": true
},
{
"why": "a zero interval",
"value": {
"poll_interval": "0s",
"label": "archie",
"dispatch": { "trigger": "assignee" }
},
"valid": false
},
{
"why": "an interval the server's parser cannot read",
"value": {
"poll_interval": "nonsense",
"label": "archie",
"dispatch": { "trigger": "assignee" }
},
"valid": false
},
{
"why": "the label trigger with no label",
"value": {
"poll_interval": "30s",
"max_retries": 3,
"dispatch": { "trigger": "label" }
},
"valid": false
},
{
"why": "the either trigger with an empty label",
"value": {
"poll_interval": "30s",
"max_retries": 0,
"label": "",
"dispatch": { "trigger": "either" }
},
"valid": false
}
],
"schedules": [
{
"why": "a single valid workflow job",
"value": [
{
"id": "daily",
"kind": "workflow",
"schedule": { "kind": "interval", "interval": "24h0m0s" }
}
],
"valid": true
},
{
"why": "a cron job with a pool the server recognises",
"value": [
{
"id": "nightly",
"kind": "workflow",
"pool": "parallel",
"schedule": { "kind": "cron", "cron": "0 3 * * *" }
}
],
"valid": true
},
{
"why": "a job with no id",
"value": [
{
"id": "",
"kind": "workflow",
"schedule": { "kind": "interval", "interval": "24h0m0s" }
}
],
"valid": false
},
{
"why": "two jobs sharing an id",
"value": [
{
"id": "daily",
"kind": "workflow",
"schedule": { "kind": "interval", "interval": "24h0m0s" }
},
{
"id": "daily",
"kind": "workflow",
"schedule": { "kind": "interval", "interval": "1h0m0s" }
}
],
"valid": false
}
]
}
}
+98
View File
@@ -0,0 +1,98 @@
package channels
import (
"context"
"log/slog"
"sync"
"time"
"github.com/samcharles93/archie-core/internal/domain/messaging"
)
// defaultEphemeralDeleteTimeout bounds one retraction. It is independent of
// the turn's own context, which may already be cancelled by the time the TTL
// elapses: the message was sent, so it still has to be removed.
const defaultEphemeralDeleteTimeout = 30 * time.Second
// SendFunc delivers one message and returns the platform's view of it, with
// at least the identifier a later retraction addresses. A sender whose
// platform cannot name a delivered message leaves ID empty, and a retraction
// of that message is then skipped rather than attempted blindly.
type SendFunc func(ctx context.Context, event messaging.MessageEvent) (messaging.MessageEvent, error)
// EphemeralSender delivers EphemeralReplies and retracts them after their TTL.
//
// The retraction runs on its own goroutine, so a TTL that has not elapsed
// never holds the turn that sent the reply. after is injected so the TTL
// fires in tests without sleeping; nil selects time.After, which is the
// production shape.
type EphemeralSender struct {
after func(time.Duration) <-chan time.Time
timeout time.Duration
log *slog.Logger
wg sync.WaitGroup
}
// NewEphemeralSender returns a sender whose retractions wait on after.
func NewEphemeralSender(log *slog.Logger, after func(time.Duration) <-chan time.Time) *EphemeralSender {
if after == nil {
after = time.After
}
if log == nil {
log = slog.Default()
}
return &EphemeralSender{
after: after,
timeout: defaultEphemeralDeleteTimeout,
log: log,
}
}
// Send delivers reply.Event through send. When sender reports the Delete
// capability, the delivered message is retracted once reply.TTL elapses;
// otherwise the reply is left in place and the fact is logged once, at debug
// level. A platform that cannot delete must never fail the turn, so the
// degrade returns the send's outcome rather than an error.
//
// A send error is returned unchanged and no retraction is scheduled: there
// is no message to remove. A non-positive TTL or a send that reported no
// message identifier also schedules nothing.
func (e *EphemeralSender) Send(ctx context.Context, sender any, reply messaging.EphemeralReply, send SendFunc) (messaging.MessageEvent, error) {
sent, err := send(ctx, reply.Event)
if err != nil {
return sent, err
}
// A reply with no lifetime, or a send the platform could not name, has
// nothing to schedule: there is either no deadline or no message to
// address.
if reply.TTL <= 0 || sent.ID == "" {
return sent, nil
}
deleter, ok := messaging.DeleterOf(sender)
if !ok {
e.log.Debug("channel cannot retract an ephemeral reply; leaving it in place",
"platform", reply.Event.Platform, "type", reply.Event.Type)
return sent, nil
}
e.wg.Go(func() {
<-e.after(reply.TTL)
// The turn's context may already be cancelled by the time the TTL
// elapses -- a delivered message is still retracted -- so the
// retraction carries its own bounded, uncancellable context.
delCtx, cancel := context.WithTimeout(context.WithoutCancel(ctx), e.timeout)
defer cancel()
if err := deleter.DeleteMessage(delCtx, sent); err != nil {
e.log.Warn("ephemeral reply retraction failed",
"platform", reply.Event.Platform, "error", err)
}
})
return sent, nil
}
// Wait blocks until every retraction started so far has finished. Tests only:
// production callers never wait on a retraction's own TTL.
func (e *EphemeralSender) Wait() {
e.wg.Wait()
}
+261
View File
@@ -0,0 +1,261 @@
package channels
import (
"bytes"
"context"
"errors"
"log/slog"
"strings"
"sync"
"testing"
"time"
"github.com/samcharles93/archie-core/internal/domain/messaging"
)
// recordingDeleter is a sender that can retract messages. report controls the
// Delete capability it advertises, so a test can prove the caller trusts the
// capability rather than the method set.
type recordingDeleter struct {
report bool
mu sync.Mutex
deleted []string
err error
}
func (d *recordingDeleter) DeleteMessage(_ context.Context, event messaging.MessageEvent) error {
d.mu.Lock()
defer d.mu.Unlock()
d.deleted = append(d.deleted, event.ID)
return d.err
}
func (d *recordingDeleter) Capabilities() messaging.AdapterCapabilities {
return messaging.AdapterCapabilities{Delete: d.report}
}
func (d *recordingDeleter) deletions() []string {
d.mu.Lock()
defer d.mu.Unlock()
return append([]string(nil), d.deleted...)
}
// plainSender is a launch-scoped sender that offers no optional delivery
// behavior at all -- the shape every adapter that cannot delete has.
type plainSender struct{}
// syncBuffer is a bytes.Buffer safe for a log record written on a retraction
// goroutine while the test reads it.
type syncBuffer struct {
mu sync.Mutex
buf bytes.Buffer
}
func (b *syncBuffer) Write(p []byte) (int, error) {
b.mu.Lock()
defer b.mu.Unlock()
return b.buf.Write(p)
}
func (b *syncBuffer) String() string {
b.mu.Lock()
defer b.mu.Unlock()
return b.buf.String()
}
func newDebugLogger(buf *syncBuffer) *slog.Logger {
return slog.New(slog.NewTextHandler(buf, &slog.HandlerOptions{Level: slog.LevelDebug}))
}
// sentEvent returns a send func that reports the platform's delivered event,
// carrying id, so a retraction has something to address.
func sentEvent(id string) SendFunc {
return func(_ context.Context, event messaging.MessageEvent) (messaging.MessageEvent, error) {
event.ID = id
return event, nil
}
}
// An ephemeral reply must be retracted once its TTL elapses, not before, and
// the retraction must address the message the send actually delivered. The
// clock is injected, so the test proves the TTL ordering without sleeping.
func TestEphemeralSenderRetractsAfterTTL(t *testing.T) {
deleter := &recordingDeleter{report: true}
fire := make(chan time.Time)
var gotTTL time.Duration
sender := NewEphemeralSender(slog.New(slog.DiscardHandler), func(d time.Duration) <-chan time.Time {
gotTTL = d
return fire
})
sent, err := sender.Send(t.Context(), deleter, messaging.EphemeralReply{
Event: messaging.MessageEvent{Type: messaging.MsgText, Text: "running tool X…"},
TTL: 5 * time.Second,
}, sentEvent("42"))
if err != nil {
t.Fatalf("Send: %v", err)
}
if sent.ID != "42" {
t.Fatalf("Send returned ID %q, want the delivered id", sent.ID)
}
if got := deleter.deletions(); len(got) != 0 {
t.Fatalf("retracted %v before the TTL elapsed, want nothing yet", got)
}
fire <- time.Now()
sender.Wait()
got := deleter.deletions()
if len(got) != 1 || got[0] != "42" {
t.Fatalf("retracted %v, want exactly the delivered message 42", got)
}
if gotTTL != 5*time.Second {
t.Fatalf("scheduled retraction after %v, want the reply's TTL 5s", gotTTL)
}
}
// A platform with no delete support must still deliver the reply: the
// message is sent, left in place, and the capability gap is logged, never
// returned as a turn error.
func TestEphemeralSenderDegradesWhenSenderCannotDelete(t *testing.T) {
var buf syncBuffer
sender := NewEphemeralSender(newDebugLogger(&buf), func(time.Duration) <-chan time.Time {
t.Error("scheduled a retraction for a sender with no delete support")
return closedTimeChan()
})
sendCalled := false
_, err := sender.Send(t.Context(), &plainSender{}, messaging.EphemeralReply{
Event: messaging.MessageEvent{Type: messaging.MsgText, Text: "thinking…"},
TTL: time.Second,
}, func(_ context.Context, event messaging.MessageEvent) (messaging.MessageEvent, error) {
sendCalled = true
event.ID = "7"
return event, nil
})
if err != nil {
t.Fatalf("Send returned an error for a sender that cannot delete: %v", err)
}
if !sendCalled {
t.Fatal("the reply was not sent; it must degrade to a normal send")
}
sender.Wait()
if !strings.Contains(buf.String(), "cannot retract") {
t.Fatalf("log = %q, want it to record the missed retraction", buf.String())
}
}
// The capability is the contract, not the method set: an adapter that
// implements DeleteMessage but reports Delete:false must not be retracted.
func TestEphemeralSenderTrustsCapabilityOverMethodSet(t *testing.T) {
deleter := &recordingDeleter{report: false}
fire := make(chan time.Time)
close(fire) // fire at once if anything schedules a retraction
sender := NewEphemeralSender(slog.New(slog.DiscardHandler), func(time.Duration) <-chan time.Time {
return fire
})
if _, err := sender.Send(t.Context(), deleter, messaging.EphemeralReply{
Event: messaging.MessageEvent{Type: messaging.MsgText, Text: "thinking…"},
TTL: time.Second,
}, sentEvent("9")); err != nil {
t.Fatalf("Send: %v", err)
}
sender.Wait()
if got := deleter.deletions(); len(got) != 0 {
t.Fatalf("retracted %v despite Capabilities().Delete == false", got)
}
}
// A failed send has no message to retract, and its error belongs to the
// caller: Send returns it unchanged and schedules nothing.
func TestEphemeralSenderPropagatesSendError(t *testing.T) {
wantErr := errors.New("send refused")
deleter := &recordingDeleter{report: true}
sender := NewEphemeralSender(slog.New(slog.DiscardHandler), func(time.Duration) <-chan time.Time {
t.Error("scheduled a retraction for a message that was never sent")
return closedTimeChan()
})
_, err := sender.Send(t.Context(), deleter, messaging.EphemeralReply{
Event: messaging.MessageEvent{Type: messaging.MsgText, Text: "thinking…"},
TTL: time.Second,
}, func(context.Context, messaging.MessageEvent) (messaging.MessageEvent, error) {
return messaging.MessageEvent{}, wantErr
})
if !errors.Is(err, wantErr) {
t.Fatalf("Send error = %v, want %v", err, wantErr)
}
sender.Wait()
if got := deleter.deletions(); len(got) != 0 {
t.Fatalf("retracted %v after a failed send", got)
}
}
// A reply with no lifetime and one whose send reported no message id both
// schedule nothing: there is either no deadline or no message to address.
func TestEphemeralSenderSchedulesNothingWithoutTTLOrMessageID(t *testing.T) {
closed := closedTimeChan()
tests := []struct {
name string
ttl time.Duration
id string
}{
{name: "zero TTL", ttl: 0, id: "11"},
{name: "negative TTL", ttl: -time.Second, id: "11"},
{name: "no message id", ttl: time.Second, id: ""},
}
for _, tt := range tests {
t.Run(tt.name, func(t *testing.T) {
deleter := &recordingDeleter{report: true}
sender := NewEphemeralSender(slog.New(slog.DiscardHandler), func(time.Duration) <-chan time.Time {
return closed
})
if _, err := sender.Send(t.Context(), deleter, messaging.EphemeralReply{
Event: messaging.MessageEvent{Type: messaging.MsgText, Text: "thinking…"},
TTL: tt.ttl,
}, sentEvent(tt.id)); err != nil {
t.Fatalf("Send: %v", err)
}
sender.Wait()
if got := deleter.deletions(); len(got) != 0 {
t.Fatalf("retracted %v, want nothing scheduled", got)
}
})
}
}
// A retraction failure is logged, not returned: the reply was delivered, and
// a platform hiccup removing it later must not surface as a turn error.
func TestEphemeralSenderLogsRetractionFailure(t *testing.T) {
var buf syncBuffer
deleter := &recordingDeleter{report: true, err: errors.New("message not found")}
fire := make(chan time.Time)
close(fire)
sender := NewEphemeralSender(newDebugLogger(&buf), func(time.Duration) <-chan time.Time {
return fire
})
if _, err := sender.Send(t.Context(), deleter, messaging.EphemeralReply{
Event: messaging.MessageEvent{Type: messaging.MsgText, Text: "thinking…"},
TTL: time.Second,
}, sentEvent("13")); err != nil {
t.Fatalf("Send: %v", err)
}
sender.Wait()
if !strings.Contains(buf.String(), "retraction failed") {
t.Fatalf("log = %q, want it to record the failed retraction", buf.String())
}
}
func closedTimeChan() <-chan time.Time {
ch := make(chan time.Time)
close(ch)
return ch
}
+13 -1
View File
@@ -54,7 +54,19 @@ type approvalResult struct {
}
// Compile-time guard.
var _ messaging.ApprovalRequester = (*telegramApprover)(nil)
var (
_ messaging.ApprovalRequester = (*telegramApprover)(nil)
_ messaging.CapabilityReporter = (*telegramApprover)(nil)
)
// Capabilities reports that this approver renders a native approval prompt.
// It claims Approval only: clarify and picker have no native Telegram UI
// wired yet, so claiming them would advertise an interaction this type
// cannot carry. A caller that sees Clarify/Picker false uses the shared text
// fallback instead of silently dropping the question.
func (a *telegramApprover) Capabilities() messaging.AdapterCapabilities {
return messaging.AdapterCapabilities{Approval: true}
}
// NewApprover returns an ApprovalRequester that renders prompts in the
// given chat. recipient is the Telegram user who may approve; in a private
@@ -413,3 +413,23 @@ func TestParseApprovalCallback(t *testing.T) {
})
}
}
// The approver renders a native approval prompt, so it must report the
// Approval capability -- and must not claim the interactive capabilities it
// has no native UI for. A caller seeing Clarify/Picker false uses the shared
// text fallback rather than dropping the question.
func TestApproverReportsApprovalCapability(t *testing.T) {
g, b, _ := newApprovalTestGateway(t)
approver := newTestApprover(g, b)
caps := messaging.CapabilitiesOf(approver)
if !caps.Approval {
t.Error("telegram approver does not report the Approval capability")
}
if caps.Clarify || caps.Picker {
t.Errorf("telegram approver over-claims interactive capabilities: %+v", caps)
}
if _, ok := messaging.ApproverOf(approver); !ok {
t.Error("ApproverOf rejected an approver that reports the Approval capability")
}
}
+60
View File
@@ -0,0 +1,60 @@
package telegram
import (
"context"
"fmt"
"strconv"
"github.com/go-telegram/bot"
"github.com/samcharles93/archie-core/internal/domain/messaging"
)
// telegramMessageDeleter removes messages the launch that built it sent,
// through the Bot API's deleteMessage method.
//
// It captures the *bot.Bot at construction rather than reading a Gateway
// field, matching telegramMediaSender and telegramApprover: a deleter belongs
// to the launch that built it, so a /restart abandons in-flight retractions
// with the outgoing bot instance instead of silently directing them through
// the new one.
type telegramMessageDeleter struct {
bot *bot.Bot
chatID int64
}
var (
_ messaging.MessageDeleter = (*telegramMessageDeleter)(nil)
_ messaging.CapabilityReporter = (*telegramMessageDeleter)(nil)
)
// NewMessageDeleter returns a MessageDeleter bound to one chat and to the bot
// instance of the launch that created it. Construct it per launch, from the
// same b that handled the update.
func (g *Gateway) NewMessageDeleter(b *bot.Bot, chatID int64) messaging.MessageDeleter {
return &telegramMessageDeleter{bot: b, chatID: chatID}
}
// Capabilities reports Delete support. Telegram can remove any message the
// bot sent in a chat it administers, so this is unconditionally true.
func (d *telegramMessageDeleter) Capabilities() messaging.AdapterCapabilities {
return messaging.AdapterCapabilities{Delete: true}
}
// DeleteMessage removes the message event names. The event's ID is the
// Telegram message ID the send returned; a non-numeric one cannot address a
// platform message, so it is reported rather than sent to the API as a
// doomed request.
func (d *telegramMessageDeleter) DeleteMessage(ctx context.Context, event messaging.MessageEvent) error {
id, err := strconv.Atoi(event.ID)
if err != nil {
return fmt.Errorf("telegram delete: message id %q is not numeric: %w", event.ID, err)
}
if _, err := d.bot.DeleteMessage(ctx, &bot.DeleteMessageParams{
ChatID: d.chatID,
MessageID: id,
}); err != nil {
return fmt.Errorf("telegram delete message %d: %w", id, err)
}
return nil
}
@@ -0,0 +1,60 @@
package telegram
import (
"context"
"fmt"
"time"
"github.com/go-telegram/bot"
"github.com/samcharles93/archie-core/internal/domain/messaging"
)
// statusNoticeTTL is how long a transient status notice stays in the chat
// before it is retracted: long enough to be read, short enough not to become
// history next to the conversation it reported on.
const statusNoticeTTL = 8 * time.Second
// sendEphemeral delivers text and retracts it after statusNoticeTTL.
//
// The retraction is best-effort by contract: a platform that cannot delete
// leaves the notice in place (channels.EphemeralSender logs the degrade), and
// a failed send is logged rather than returned, because a status notice must
// never take the operation it narrates down with it.
func (g *Gateway) sendEphemeral(ctx context.Context, b *bot.Bot, chatID int64, messageThreadID int, text string) {
sender := g.newEphemeralSender()
_, err := sender.Send(ctx, g.NewMessageDeleter(b, chatID), messaging.EphemeralReply{
Event: messaging.MessageEvent{
Type: messaging.MsgText,
Text: text,
ChannelID: fmt.Sprintf("%d", chatID),
Platform: "telegram",
},
TTL: statusNoticeTTL,
}, func(ctx context.Context, event messaging.MessageEvent) (messaging.MessageEvent, error) {
id := g.sendTextMessage(ctx, b, chatID, messageThreadID, event.Text)
if id == 0 {
// Nothing was delivered, so there is nothing to name or retract.
return messaging.MessageEvent{}, nil
}
event.ID = fmt.Sprintf("%d", id)
return event, nil
})
if err != nil {
g.log.Error("ephemeral notice send failed", "error", err)
}
}
// sendTextMessage delivers one message and returns its Bot API message ID.
//
// It returns 0 for text that must be split across messages: a retraction
// addresses exactly one message, so a reply that does not fit in one is left
// intact rather than half-deleted.
func (g *Gateway) sendTextMessage(ctx context.Context, b *bot.Bot, chatID int64, messageThreadID int, text string) int {
blocks := markdownToBlocks(text)
if blockTextLen(blocks) > messageMaxLen {
g.sendMessage(ctx, b, chatID, messageThreadID, text)
return 0
}
return g.sendBlocks(ctx, b, chatID, messageThreadID, blocks, "send ephemeral message failed")
}
@@ -0,0 +1,160 @@
package telegram
import (
"log/slog"
"net/http"
"net/http/httptest"
"strings"
"testing"
"time"
"github.com/go-telegram/bot"
"github.com/samcharles93/archie-core/internal/channels"
"github.com/samcharles93/archie-core/internal/domain/messaging"
)
func newTestDeleter(t *testing.T, serverURL string, chatID int64) messaging.MessageDeleter {
t.Helper()
b, err := bot.New("1:test", bot.WithServerURL(serverURL), bot.WithSkipGetMe())
if err != nil {
t.Fatalf("new test bot: %v", err)
}
return New("1:test", []int64{42}, slog.New(slog.DiscardHandler)).NewMessageDeleter(b, chatID)
}
// The delete capability must be reported, so an ephemeral reply's retraction
// path can be chosen without attempting a doomed call.
func TestMessageDeleterReportsDeleteCapability(t *testing.T) {
_, srv := newMediaAPI()
defer srv.Close()
deleter := newTestDeleter(t, srv.URL, 555)
if caps := messaging.CapabilitiesOf(deleter); !caps.Delete {
t.Error("telegram message deleter does not report the Delete capability")
}
}
// A retraction must reach the Bot API's deleteMessage method with the chat and
// the message the send returned.
func TestMessageDeleterCallsDeleteMessage(t *testing.T) {
api, srv := newMediaAPI()
api.body = `{"ok":true,"result":true}`
defer srv.Close()
deleter := newTestDeleter(t, srv.URL, 555)
if err := deleter.DeleteMessage(t.Context(), messaging.MessageEvent{ID: "77"}); err != nil {
t.Fatalf("DeleteMessage: %v", err)
}
method, form := api.called()
if method != "deleteMessage" {
t.Fatalf("Bot API method = %q, want deleteMessage", method)
}
if form["message_id"] != "77" {
t.Errorf("message_id = %q, want 77", form["message_id"])
}
if form["chat_id"] != "555" {
t.Errorf("chat_id = %q, want the bound chat 555", form["chat_id"])
}
}
// A message ID the platform never issued cannot address a message, so it is
// reported before any request rather than sent as a doomed one.
func TestMessageDeleterRejectsNonNumericID(t *testing.T) {
api, srv := newMediaAPI()
api.body = `{"ok":true,"result":true}`
defer srv.Close()
deleter := newTestDeleter(t, srv.URL, 555)
if err := deleter.DeleteMessage(t.Context(), messaging.MessageEvent{ID: "not-a-number"}); err == nil {
t.Fatal("DeleteMessage accepted a non-numeric message ID")
}
if method, _ := api.called(); method != "" {
t.Fatalf("Bot API method = %q, want no request for an undeliverable ID", method)
}
}
// An API failure must surface as an error so the caller can log it; a failed
// retraction is never silent.
func TestMessageDeleterReportsAPIError(t *testing.T) {
api, srv := newMediaAPI()
api.body = `{"ok":false,"error_code":400,"description":"message to delete not found"}`
defer srv.Close()
deleter := newTestDeleter(t, srv.URL, 555)
if err := deleter.DeleteMessage(t.Context(), messaging.MessageEvent{ID: "404"}); err == nil {
t.Fatal("DeleteMessage reported success for a rejected request")
}
}
// sendEphemeral is the transport honouring an ephemeral reply end to end: it
// sends the status notice, then retracts the message the send returned once
// the injected clock's TTL fires. The injected sender keeps the test off a
// real eight-second wait.
func TestSendEphemeralRetractsStatusNotice(t *testing.T) {
type call struct {
method string
form map[string]string
}
calls := make(chan call, 8)
srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
parts := strings.Split(strings.Trim(r.URL.Path, "/"), "/")
method := parts[len(parts)-1]
_ = r.ParseMultipartForm(1 << 20)
form := map[string]string{}
for k, v := range r.Form {
if len(v) > 0 {
form[k] = v[0]
}
}
calls <- call{method: method, form: form}
w.Header().Set("Content-Type", "application/json")
if method == "deleteMessage" {
_, _ = w.Write([]byte(`{"ok":true,"result":true}`))
return
}
_, _ = w.Write([]byte(`{"ok":true,"result":{"message_id":321}}`))
}))
defer srv.Close()
b, err := bot.New("1:test", bot.WithServerURL(srv.URL), bot.WithSkipGetMe())
if err != nil {
t.Fatalf("new test bot: %v", err)
}
g := New("1:test", []int64{42}, slog.New(slog.DiscardHandler))
g.newEphemeralSender = func() *channels.EphemeralSender {
return channels.NewEphemeralSender(g.log, func(time.Duration) <-chan time.Time {
ch := make(chan time.Time)
close(ch)
return ch
})
}
g.sendEphemeral(t.Context(), b, 555, 0, "✅ Archie reloaded.")
var sent, retracted *call
deadline := time.After(2 * time.Second)
for sent == nil || retracted == nil {
select {
case c := <-calls:
switch c.method {
case "sendRichMessage", "sendMessage":
sent = &c
case "deleteMessage":
retracted = &c
}
case <-deadline:
t.Fatalf("timed out; send=%v retraction=%v", sent, retracted)
}
}
if retracted.form["message_id"] != "321" {
t.Errorf("retracted message_id = %q, want the sent message 321", retracted.form["message_id"])
}
if retracted.form["chat_id"] != "555" {
t.Errorf("retraction chat_id = %q, want the chat the notice was sent to", retracted.form["chat_id"])
}
}
+2 -2
View File
@@ -82,10 +82,10 @@ func (f *fakeChatContract) SetPersona(ctx context.Context, id, name string) (boo
return true, nil
}
func (f *fakeChatContract) ApplyTaskAction(ctx context.Context, identity string, taskID int64, action taskstate.Action) (messaging.TaskActionResult, error) {
func (f *fakeChatContract) ApplyTaskAction(ctx context.Context, identity string, taskID int64, action taskstate.Action, _ taskactions.ActionPayload) (messaging.TaskActionResult, error) {
return messaging.TaskActionResult{}, nil
}
func (f *fakeChatContract) ApplyOperatorTaskAction(ctx context.Context, actor taskactions.Actor, taskID int64, action taskstate.Action) (messaging.TaskActionResult, error) {
func (f *fakeChatContract) ApplyOperatorTaskAction(ctx context.Context, actor taskactions.Actor, taskID int64, action taskstate.Action, _ taskactions.ActionPayload) (messaging.TaskActionResult, error) {
return messaging.TaskActionResult{}, nil
}
+13 -12
View File
@@ -40,42 +40,42 @@ func extractInboundMedia(msg *models.Message) (inboundMedia, bool) {
}
}
}
return inboundMedia{att, "[photo]"}, true
return inboundMedia{attachment: att, note: "[photo]"}, true
case msg.Document != nil:
return inboundMedia{messaging.MediaAttachment{
return inboundMedia{attachment: messaging.MediaAttachment{
Type: "document",
FileID: msg.Document.FileID,
MIMEType: msg.Document.MimeType,
FileName: msg.Document.FileName,
FileSize: nilIfZero(msg.Document.FileSize),
}, documentNote(msg.Document.FileName)}, true
}, note: documentNote(msg.Document.FileName)}, true
case msg.Video != nil:
return inboundMedia{videoAttachment(msg.Video.FileID, msg.Video.MimeType, nilIfZero(msg.Video.FileSize), &msg.Video.Width, &msg.Video.Height, &msg.Video.Duration), "[video]"}, true
return inboundMedia{attachment: videoAttachment(msg.Video.FileID, msg.Video.MimeType, nilIfZero(msg.Video.FileSize), &msg.Video.Width, &msg.Video.Height, &msg.Video.Duration), note: "[video]"}, true
case msg.VideoNote != nil:
size := int64(msg.VideoNote.FileSize)
return inboundMedia{videoAttachment(msg.VideoNote.FileID, "video/mp4", nilIfZero(size), nil, nil, &msg.VideoNote.Duration), "[video message]"}, true
return inboundMedia{attachment: videoAttachment(msg.VideoNote.FileID, "video/mp4", nilIfZero(size), nil, nil, &msg.VideoNote.Duration), note: "[video message]"}, true
case msg.Audio != nil:
return inboundMedia{messaging.MediaAttachment{
return inboundMedia{attachment: messaging.MediaAttachment{
Type: "audio",
FileID: msg.Audio.FileID,
MIMEType: msg.Audio.MimeType,
FileName: msg.Audio.FileName,
FileSize: nilIfZero(msg.Audio.FileSize),
Duration: &msg.Audio.Duration,
}, "[audio]"}, true
}, note: "[audio]"}, true
case msg.Voice != nil:
return inboundMedia{messaging.MediaAttachment{
Type: "audio",
return inboundMedia{attachment: messaging.MediaAttachment{
Type: messaging.MediaTypeVoice,
FileID: msg.Voice.FileID,
MIMEType: msg.Voice.MimeType,
FileSize: nilIfZero(msg.Voice.FileSize),
Duration: &msg.Voice.Duration,
}, "[voice message]"}, true
}, note: "[voice message]"}, true
}
return inboundMedia{}, false
}
@@ -114,8 +114,9 @@ func nilIfZero(v int64) *int64 {
// maxInboundDownloadBytes is the Bot API download ceiling: bots cannot
// fetch files larger than 20 MB through getFile, so a larger attachment
// must be refused with a clear notice rather than met with a Bot API 400
// nobody reads.
var maxInboundDownloadBytes int64 = 20 * 1024 * 1024
// nobody reads. It is the messaging contract's inbound attachment ceiling,
// so the value the Gateway must be able to receive stays in one place.
var maxInboundDownloadBytes int64 = messaging.MaxInboundAttachmentBytes
var errInboundMediaTooLarge = errors.New("file exceeds the bot download limit")
+102 -1
View File
@@ -158,7 +158,7 @@ func TestExtractInboundMediaMapsTelegramKinds(t *testing.T) {
msg: &models.Message{
Voice: &models.Voice{FileID: "voice1", MimeType: "audio/ogg", Duration: 8},
},
wantType: "audio",
wantType: messaging.MediaTypeVoice,
wantFileID: "voice1",
wantMIME: "audio/ogg",
wantNote: "[voice message]",
@@ -333,6 +333,107 @@ func TestDefaultHandlerSilentForNonContentMessages(t *testing.T) {
}
}
// voiceUpdate is a Telegram voice note from the authorized sender.
func voiceUpdate(fileID string) *models.Message {
return &models.Message{
ID: 11,
Date: 1,
From: &models.User{ID: 42, Username: "sam"},
Chat: models.Chat{ID: 7, Type: "private"},
Voice: &models.Voice{FileID: fileID, MimeType: "audio/ogg", Duration: 8},
}
}
// TestDefaultHandlerMarksVoiceNoteAsSpeech drives the frontend half of the
// boundary: the messaging process downloads the note, classifies it as speech
// (messaging.MediaTypeVoice), and carries the bytes and the note to the turn.
// Turning those bytes into text is the model-owning process's job -- this
// process holds no provider credential -- so no transcription happens here.
func TestDefaultHandlerMarksVoiceNoteAsSpeech(t *testing.T) {
voiceBytes := []byte("ogg-voice-bytes")
b, _ := newRecordedAPI(t, func(w http.ResponseWriter, r *http.Request) bool {
if strings.HasPrefix(r.URL.Path, "/file/bot1:test/") {
_, _ = w.Write(voiceBytes)
return true
}
return false
})
inboundCh := make(chan messaging.Inbound, 1)
client := &fakeChatContract{
streamFunc: func(_ context.Context, in messaging.Inbound) (<-chan messaging.ChatEvent, error) {
inboundCh <- in
ch := make(chan messaging.ChatEvent, 1)
ch <- messaging.ChatEvent{Kind: "done", Text: "done", SessionID: "s1"}
close(ch)
return ch, nil
},
}
g := testMediaGateway(b)
g.defaultHandler(client)(context.Background(), b, &models.Update{Message: voiceUpdate("voice1")})
select {
case in := <-inboundCh:
if len(in.Media) != 1 || in.Media[0].Type != messaging.MediaTypeVoice {
t.Fatalf("inbound media = %#v, want the voice attachment marked as speech", in.Media)
}
if string(in.Media[0].Data) != string(voiceBytes) {
t.Errorf("inbound media bytes = %q, want the downloaded voice bytes", in.Media[0].Data)
}
if in.Message.Text != "[voice message]" {
t.Errorf("Text = %q, want the media note preserved for the model-owning side", in.Message.Text)
}
case <-time.After(5 * time.Second):
t.Fatal("voice note never reached the chat turn")
}
}
// TestDefaultHandlerMarksPlainAudioAsAudio keeps the speech distinction: a
// forwarded music file is ordinary audio, so the model-owning side knows not
// to spend a transcription call on it.
func TestDefaultHandlerMarksPlainAudioAsAudio(t *testing.T) {
b, _ := newRecordedAPI(t, func(w http.ResponseWriter, r *http.Request) bool {
if strings.HasPrefix(r.URL.Path, "/file/bot1:test/") {
_, _ = w.Write([]byte("mp3-bytes"))
return true
}
return false
})
inboundCh := make(chan messaging.Inbound, 1)
client := &fakeChatContract{
streamFunc: func(_ context.Context, in messaging.Inbound) (<-chan messaging.ChatEvent, error) {
inboundCh <- in
ch := make(chan messaging.ChatEvent, 1)
ch <- messaging.ChatEvent{Kind: "done", Text: "done", SessionID: "s1"}
close(ch)
return ch, nil
},
}
g := testMediaGateway(b)
g.defaultHandler(client)(context.Background(), b, &models.Update{Message: &models.Message{
ID: 12,
Date: 1,
From: &models.User{ID: 42, Username: "sam"},
Chat: models.Chat{ID: 7, Type: "private"},
Audio: &models.Audio{FileID: "aud1", FileName: "song.mp3", MimeType: "audio/mpeg", Duration: 200},
}})
select {
case in := <-inboundCh:
if len(in.Media) != 1 || in.Media[0].Type != messaging.MediaTypeAudio {
t.Fatalf("inbound media = %#v, want the plain audio attachment", in.Media)
}
if in.Message.Text != "[audio]" {
t.Errorf("Text = %q, want the plain audio note", in.Message.Text)
}
case <-time.After(5 * time.Second):
t.Fatal("audio update never reached the chat turn")
}
}
// TestDownloadMediaCoversSuccessAndLimits exercises the download the turn
// relies on: the Bot API getFile lookup, the file fetch, and the size gate.
func TestDownloadMediaCoversSuccessAndLimits(t *testing.T) {
+217
View File
@@ -0,0 +1,217 @@
package telegram
import (
"context"
"fmt"
"strings"
"time"
"github.com/go-telegram/bot"
"github.com/go-telegram/bot/models"
"github.com/samcharles93/archie-core/internal/domain/messaging"
)
// interactiveReplyTimeout bounds how long one clarify or picker request waits
// for the human's reply. It matches the approval window so the two blocking
// interactions expire together and neither can wedge a turn indefinitely.
const interactiveReplyTimeout = messaging.ToolApprovalTimeout
// pendingReplyKey identifies the chat a blocked clarify/picker prompt belongs
// to, so only the addressed human's next text is consumed as the answer.
type pendingReplyKey struct {
chatID int64
threadID int
recipient int64
}
// pendingReply is one clarify/picker prompt waiting on the human's next text.
// It is the text-reply analogue of pendingApproval, kept separate so the
// button (callback) and message (text) models cannot entangle.
type pendingReply struct {
resultCh chan replyResult
expiresAt time.Time
}
// replyResult carries the human's typed answer (or the wait's error) back to
// the blocked interaction.
type replyResult struct {
text string
err error
}
// telegramInteractor carries clarify and picker interactions for one chat on
// the channel's shared text fallback. Telegram renders no generic question or
// single-choice widget in the agent-turn path, so both interactions appear as
// text the human answers by replying; reporting the Clarify and Picker
// capabilities is therefore honest, because the adapter can carry the
// interaction even though it costs a typed answer.
type telegramInteractor struct {
gw *Gateway
bot *bot.Bot
chatID int64
threadID int
recipient int64
}
// Compile-time guards: the adapter carries both interactions and reports them
// through the capability contract, so ClarifierOf/PickerOf resolve it.
var (
_ messaging.ClarifyRequester = (*telegramInteractor)(nil)
_ messaging.PickerRequester = (*telegramInteractor)(nil)
_ messaging.CapabilityReporter = (*telegramInteractor)(nil)
)
// Capabilities reports that this adapter can carry a clarify question and a
// single-choice picker. Both go through the shared text fallback rather than a
// native widget, and the capability is about carrying the interaction, not
// about the widget, so claiming them is what keeps a question from being
// silently dropped.
func (i *telegramInteractor) Capabilities() messaging.AdapterCapabilities {
return messaging.AdapterCapabilities{Clarify: true, Picker: true}
}
// NewInteractor returns the channel's clarify/picker adapter for one chat,
// already resolved to the requesters its capability report advertises. The
// gateway turn stores it with WithInteractive so its question tool can reach
// the human.
func (g *Gateway) NewInteractor(b *bot.Bot, chatID int64, threadID int, recipient int64) messaging.Interactive {
i := &telegramInteractor{
gw: g,
bot: b,
chatID: chatID,
threadID: threadID,
recipient: recipient,
}
return messaging.InteractiveOf(i)
}
// RequestClarification poses the question as text and blocks for the reply.
func (i *telegramInteractor) RequestClarification(ctx context.Context, req messaging.ClarifyRequest) (string, error) {
ctx, cancel := context.WithTimeout(ctx, interactiveReplyTimeout)
defer cancel()
fallback := i.fallback()
return fallback.RequestClarification(ctx, req)
}
// RequestChoice poses the options as numbered text and matches the typed
// reply. The domain's text fallback owns the rendering and matching, so the
// degrade is one implementation rather than a second parser here.
func (i *telegramInteractor) RequestChoice(ctx context.Context, req messaging.PickerRequest) (messaging.InteractiveChoice, error) {
ctx, cancel := context.WithTimeout(ctx, interactiveReplyTimeout)
defer cancel()
fallback := i.fallback()
return fallback.RequestChoice(ctx, req)
}
// fallback builds a fresh text fallback for one interaction. Registering the
// pending reply inside Send, before the prompt is delivered, closes the race
// where a fast reply arrives before the blocked call is listening.
func (i *telegramInteractor) fallback() messaging.TextFallback {
key := pendingReplyKey{chatID: i.chatID, threadID: i.threadID, recipient: i.recipient}
ch := make(chan replyResult, 1)
return messaging.TextFallback{
Send: func(ctx context.Context, text string) error {
i.gw.registerPendingReply(key, pendingReply{
resultCh: ch,
expiresAt: time.Now().Add(interactiveReplyTimeout),
})
if err := i.send(ctx, text); err != nil {
i.gw.removePendingReply(key)
return err
}
return nil
},
Reply: func(ctx context.Context) (string, error) {
select {
case res := <-ch:
return res.text, res.err
case <-ctx.Done():
i.gw.removePendingReply(key)
return "", ctx.Err()
}
},
}
}
// send delivers one prompt. ForceReply makes the reply affordance explicit in
// a chat with other traffic, so the human answers the question rather than
// sending a message the turn cannot tell apart from a new request.
func (i *telegramInteractor) send(ctx context.Context, text string) error {
params := &bot.SendMessageParams{
ChatID: i.chatID,
Text: text,
ReplyMarkup: &models.ForceReply{ForceReply: true, Selective: true},
}
if i.threadID != 0 {
params.MessageThreadID = i.threadID
}
if _, err := i.bot.SendMessage(ctx, params); err != nil {
return fmt.Errorf("send interactive prompt: %w", err)
}
return nil
}
// registerPendingReply stores a pending interaction, pruning any that have
// already expired.
func (g *Gateway) registerPendingReply(key pendingReplyKey, pr pendingReply) {
g.interactiveMu.Lock()
defer g.interactiveMu.Unlock()
now := time.Now()
for existing, pending := range g.pendingReplies {
if !pending.expiresAt.After(now) {
delete(g.pendingReplies, existing)
}
}
g.pendingReplies[key] = &pr
}
// removePendingReply drops a pending interaction, if present.
func (g *Gateway) removePendingReply(key pendingReplyKey) {
g.interactiveMu.Lock()
defer g.interactiveMu.Unlock()
delete(g.pendingReplies, key)
}
// consumePendingReply atomically claims the pending interaction for the chat
// and delivers the typed answer. It rejects expired entries and reports
// whether the text was consumed as an answer.
func (g *Gateway) consumePendingReply(key pendingReplyKey, text string) bool {
g.interactiveMu.Lock()
pr, found := g.pendingReplies[key]
expired := !found || !pr.expiresAt.After(time.Now())
if !expired {
delete(g.pendingReplies, key)
}
g.interactiveMu.Unlock()
if expired {
return false
}
select {
case pr.resultCh <- replyResult{text: text}:
return true
default:
// The channel already has a result (the wait timed out between the
// expiry check and now); do not report a reply that was not taken.
return false
}
}
// deliverInteractiveReply routes a text reply to a blocked clarify/picker
// request when one is pending for this chat and sender, and reports whether
// the message was consumed. A consumed reply is not dispatched as a new turn;
// a command is never consumed, so /stop still cancels the blocked turn.
func (g *Gateway) deliverInteractiveReply(msg *models.Message) bool {
if msg == nil || msg.From == nil {
return false
}
text := strings.TrimSpace(msg.Text)
if text == "" || strings.HasPrefix(text, "/") {
return false
}
return g.consumePendingReply(pendingReplyKey{
chatID: msg.Chat.ID,
threadID: msg.MessageThreadID,
recipient: msg.From.ID,
}, text)
}
@@ -0,0 +1,168 @@
package telegram
import (
"context"
"strings"
"testing"
"time"
"github.com/go-telegram/bot/models"
"github.com/samcharles93/archie-core/internal/domain/messaging"
)
// waitForPendingReply polls the gateway's pending-reply map until a blocked
// interaction has registered itself. Polling the guarded map is race-free:
// the register happens-before the requester blocks, and the map is guarded by
// interactiveMu.
func waitForPendingReply(t *testing.T, g *Gateway) {
t.Helper()
deadline := time.Now().Add(2 * time.Second)
for time.Now().Before(deadline) {
g.interactiveMu.Lock()
n := len(g.pendingReplies)
g.interactiveMu.Unlock()
if n > 0 {
return
}
time.Sleep(5 * time.Millisecond)
}
t.Fatal("no pending interactive reply was registered")
}
func interactiveReplyMessage(chatID, recipient int64, text string) *models.Message {
return &models.Message{
From: &models.User{ID: recipient},
Chat: models.Chat{ID: chatID, Type: models.ChatTypePrivate},
Text: text,
}
}
// sentPrompt returns the text of the first sendMessage whose body contains
// want, so a test can assert what the adapter actually rendered.
func sentPrompt(requests []telegramRequest, want string) string {
for _, request := range requests {
if request.method == "sendMessage" && strings.Contains(request.form["text"], want) {
return request.form["text"]
}
}
return ""
}
func TestInteractorReportsClarifyAndPicker(t *testing.T) {
g, b, _ := newApprovalTestGateway(t)
interactor := g.NewInteractor(b, approvalTestChatID, 0, approvalTestRecipient)
if interactor.Clarifier == nil {
t.Error("interactor does not report the Clarify capability")
}
if interactor.Picker == nil {
t.Error("interactor does not report the Picker capability")
}
}
// TestInteractivePickUsesTextFallback is the acceptance test for step 4 at the
// adapter: Telegram has no generic picker widget in the turn path, so the
// choice must be rendered as numbered text and matched from the human's typed
// reply rather than failing or being dropped.
func TestInteractivePickUsesTextFallback(t *testing.T) {
g, b, requests := newApprovalTestGateway(t)
interactor := g.NewInteractor(b, approvalTestChatID, 0, approvalTestRecipient)
type outcome struct {
choice messaging.InteractiveChoice
err error
}
done := make(chan outcome, 1)
go func() {
choice, err := interactor.Picker.RequestChoice(context.Background(), messaging.PickerRequest{
Prompt: "Which branch?",
Options: []messaging.InteractiveChoice{
{ID: "main", Label: "main"},
{ID: "release", Label: "release"},
},
})
done <- outcome{choice: choice, err: err}
}()
waitForPendingReply(t, g)
if !g.deliverInteractiveReply(interactiveReplyMessage(approvalTestChatID, approvalTestRecipient, "2")) {
t.Fatal("deliverInteractiveReply did not consume the typed reply")
}
select {
case got := <-done:
if got.err != nil {
t.Fatalf("RequestChoice error = %v", got.err)
}
if got.choice.ID != "release" {
t.Errorf("choice = %+v, want release", got.choice)
}
case <-time.After(2 * time.Second):
t.Fatal("RequestChoice did not return after the reply was delivered")
}
if prompt := sentPrompt(*requests, "Which branch?"); !strings.Contains(prompt, "1) main") || !strings.Contains(prompt, "2) release") {
t.Errorf("picker prompt was not rendered as numbered text: %q", prompt)
}
}
func TestInteractiveClarifyReturnsTypedAnswer(t *testing.T) {
g, b, _ := newApprovalTestGateway(t)
interactor := g.NewInteractor(b, approvalTestChatID, 0, approvalTestRecipient)
type outcome struct {
answer string
err error
}
done := make(chan outcome, 1)
go func() {
answer, err := interactor.Clarifier.RequestClarification(context.Background(), messaging.ClarifyRequest{Question: "Which environment?"})
done <- outcome{answer: answer, err: err}
}()
waitForPendingReply(t, g)
if !g.deliverInteractiveReply(interactiveReplyMessage(approvalTestChatID, approvalTestRecipient, " staging ")) {
t.Fatal("deliverInteractiveReply did not consume the typed reply")
}
select {
case got := <-done:
if got.err != nil {
t.Fatalf("RequestClarification error = %v", got.err)
}
if got.answer != "staging" {
t.Errorf("answer = %q, want staging (trimmed)", got.answer)
}
case <-time.After(2 * time.Second):
t.Fatal("RequestClarification did not return after the reply was delivered")
}
}
// A command typed while a prompt is pending must reach the command handler, not
// be swallowed as the answer: /stop is how the human cancels the blocked turn.
func TestDeliverInteractiveReplyIgnoresCommands(t *testing.T) {
g, b, _ := newApprovalTestGateway(t)
interactor := g.NewInteractor(b, approvalTestChatID, 0, approvalTestRecipient)
ctx, cancel := context.WithTimeout(context.Background(), 200*time.Millisecond)
defer cancel()
done := make(chan error, 1)
go func() {
_, err := interactor.Clarifier.RequestClarification(ctx, messaging.ClarifyRequest{Question: "Which environment?"})
done <- err
}()
waitForPendingReply(t, g)
if g.deliverInteractiveReply(interactiveReplyMessage(approvalTestChatID, approvalTestRecipient, "/stop")) {
t.Error("deliverInteractiveReply consumed a command as the answer")
}
select {
case err := <-done:
if err == nil {
t.Error("RequestClarification returned nil error after its context expired")
}
case <-time.After(2 * time.Second):
t.Fatal("RequestClarification did not return after its context expired")
}
}
+47 -10
View File
@@ -107,6 +107,13 @@ type Gateway struct {
approvalMu sync.Mutex
pendingApprovals map[string]*pendingApproval
// interactive state (gateway clarify/picker). A blocked clarify or
// picker waits on one entry here, keyed by the chat it was posed to, so
// the next text from that human is consumed as the answer rather than
// dispatched as a new turn.
interactiveMu sync.Mutex
pendingReplies map[pendingReplyKey]*pendingReply
// turns serialises chat turns per session off the update worker and
// makes the running one cancellable by /stop. Rebuilt on every launch
// so a restart abandons in-flight turns with the old bot instance.
@@ -142,6 +149,12 @@ type Gateway struct {
// where the library's default endpoint is used. It exists so launch's
// wiring can be tested without reaching api.telegram.org.
serverURL string
// newEphemeralSender builds the retracting sender sendEphemeral uses.
// Defaults to channels.NewEphemeralSender; tests override it to inject a
// clock so a status notice's retraction fires without waiting out its
// TTL, matching liveReply's newMediaSender seam.
newEphemeralSender func() *channels.EphemeralSender
}
type UpdateService interface {
@@ -161,7 +174,7 @@ type ReleaseAnnouncer interface {
// New returns an unstarted Gateway. Call Start to begin long-polling.
func New(token string, allowedUserIDs []int64, log *slog.Logger) *Gateway {
return &Gateway{
g := &Gateway{
Token: token,
AllowedUserIDs: allowedUserIDs,
restartCh: make(chan restartRequest, 1),
@@ -172,8 +185,13 @@ func New(token string, allowedUserIDs []int64, log *slog.Logger) *Gateway {
dangerousActions: make(map[string]dangerousAction),
permanentApprovals: nil,
pendingApprovals: make(map[string]*pendingApproval),
pendingReplies: make(map[pendingReplyKey]*pendingReply),
log: log.With("component", "gateway-telegram"),
}
g.newEphemeralSender = func() *channels.EphemeralSender {
return channels.NewEphemeralSender(g.log, nil)
}
return g
}
func (g *Gateway) Name() string { return "telegram" }
@@ -221,8 +239,12 @@ func (g *Gateway) Start(ctx context.Context, client messaging.ChatContract, life
return err
}
// Confirm to whoever asked, now that the new instance can send.
// It is a status notice, not a record: it retracts itself after a
// short TTL so it does not sit in the chat as a stale "reloaded"
// line. The pre-restart "Reloading…" message from the outgoing
// instance stays as the durable acknowledgement.
if req := g.pendingRestart; req != nil && req.chatID != 0 {
g.sendMessage(runCtx, b, req.chatID, req.threadID, "✅ Archie reloaded.")
g.sendEphemeral(runCtx, b, req.chatID, req.threadID, "✅ Archie reloaded.")
}
if g.pendingRestart != nil {
g.pendingRestart = nil
@@ -532,6 +554,11 @@ func (g *Gateway) defaultHandler(client messaging.ChatContract) bot.HandlerFunc
if !ok {
return
}
// A text reply to a pending clarify/picker prompt is consumed here so
// it answers the blocked interaction instead of starting a new turn.
if g.deliverInteractiveReply(msg) {
return
}
// Media messages arrive before the text gates: a photo has no text,
// so checking it first is what keeps attachments from being dropped.
if _, isMedia := extractInboundMedia(msg); isMedia {
@@ -600,8 +627,9 @@ func (g *Gateway) handleCallback(ctx context.Context, b *bot.Bot, update *models
// turnInbound renders a Telegram message as the channel-neutral inbound a
// chat turn is built from, with media attachments attached. ok is false
// for messages with neither text nor media, which are not turns at all.
func turnInbound(msg *models.Message) (messaging.Inbound, bool) {
media, hasMedia := extractInboundMedia(msg)
// media/hasMedia are the caller's already-extracted attachment, so the
// caller keeps the voice-note flag it needs without extracting twice.
func turnInbound(msg *models.Message, media inboundMedia, hasMedia bool) (messaging.Inbound, bool) {
if !hasMedia && msg.Text == "" {
return messaging.Inbound{}, false
}
@@ -629,7 +657,8 @@ func turnInbound(msg *models.Message) (messaging.Inbound, bool) {
}
func (g *Gateway) submitTurn(ctx context.Context, b *bot.Bot, msg *models.Message, client messaging.ChatContract) {
gm, isTurn := turnInbound(msg)
media, hasMedia := extractInboundMedia(msg)
gm, isTurn := turnInbound(msg, media, hasMedia)
if !isTurn {
return
}
@@ -649,6 +678,10 @@ func (g *Gateway) submitTurn(ctx context.Context, b *bot.Bot, msg *models.Messag
approver := g.NewApprover(b, chatID, threadID, msg.From.ID)
turnCtx = messaging.WithApprovalRequester(turnCtx, approver)
// A clarify or picker prompt blocks the turn on this adapter; when the
// channel cannot carry one the question tool is simply not offered.
turnCtx = messaging.WithInteractive(turnCtx, g.NewInteractor(b, chatID, threadID, msg.From.ID))
// reply appears as it is written; the typing indicator covers the
// gap before the first token and any non-streaming path.
stopTyping := g.startTyping(turnCtx, b, chatID, threadID)
@@ -865,12 +898,14 @@ func (g *Gateway) sendMessage(ctx context.Context, b *bot.Bot, chatID int64, mes
}
// sendBlocks delivers one message as a Telegram rich message built from
// structured blocks.
// structured blocks and returns the Bot API message ID, or 0 when nothing was
// delivered. The ID is what an EphemeralReply's retraction addresses; callers
// that only send ignore it.
//
// Rich messages are a recent Bot API addition, so a rejection here is treated
// as "unsupported" rather than fatal: fall back to sending the blocks as plain
// text so the user still receives the reply, unformatted, instead of silence.
func (g *Gateway) sendBlocks(ctx context.Context, b *bot.Bot, chatID int64, messageThreadID int, blocks []models.InputRichBlock, errMsg string) {
func (g *Gateway) sendBlocks(ctx context.Context, b *bot.Bot, chatID int64, messageThreadID int, blocks []models.InputRichBlock, errMsg string) int {
rich := &bot.SendRichMessageParams{
ChatID: chatID,
RichMessage: models.InputRichMessage{Blocks: blocks},
@@ -878,9 +913,9 @@ func (g *Gateway) sendBlocks(ctx context.Context, b *bot.Bot, chatID int64, mess
if messageThreadID != 0 {
rich.MessageThreadID = messageThreadID
}
if _, err := b.SendRichMessage(ctx, rich); err == nil {
if msg, err := b.SendRichMessage(ctx, rich); err == nil {
g.log.Debug("telegram message sent", "renderer", telegramRenderVersion, "mode", "rich")
return
return msg.ID
} else {
g.log.Warn("rich send failed, retrying unformatted", "error", err)
}
@@ -889,10 +924,12 @@ func (g *Gateway) sendBlocks(ctx context.Context, b *bot.Bot, chatID int64, mess
if messageThreadID != 0 {
plain.MessageThreadID = messageThreadID
}
if _, err := b.SendMessage(ctx, plain); err != nil {
if msg, err := b.SendMessage(ctx, plain); err != nil {
g.log.Error(errMsg, "error", err)
return 0
} else {
g.log.Warn("telegram message sent via plain fallback", "renderer", telegramRenderVersion, "mode", "plain")
return msg.ID
}
}
+71
View File
@@ -140,6 +140,77 @@ func TestReportPendingUpdateSendsAndClearsReport(t *testing.T) {
}
}
// TestReportPendingUpdateVerifiesAgainstRunningVersions drives the acceptance
// both ways through the real relay: a wired RunningVersions must reach the
// message the operator reads, confirming a claim this process can vouch for and
// contradicting one it cannot. Without the wiring the same report relays as
// "I still can't confirm which version" on every success, which is the bug this
// covers.
func TestReportPendingUpdateVerifiesAgainstRunningVersions(t *testing.T) {
const (
daemon = releaseupdate.ComponentDaemon
previous = "1.0.0"
installed = "1.2.3"
)
tests := []struct {
name string
installed string
wantText []string
denyText []string
}{
{
name: "confirmed when the running version matches the claim",
installed: installed,
wantText: []string{"update complete"},
denyText: []string{"can't confirm", "did not take effect"},
},
{
name: "contradicted when the running version differs from the claim",
installed: "9.9.9",
wantText: []string{"did not take effect", installed},
denyText: []string{"update complete", "can't confirm"},
},
}
for _, test := range tests {
t.Run(test.name, func(t *testing.T) {
path := filepath.Join(t.TempDir(), "update-report.json")
report := releaseupdate.Report{
Channel: "telegram", ChatID: 7, ThreadID: 3,
Previous: map[string]string{daemon: previous},
Installed: map[string]string{daemon: test.installed},
HealthCheck: "passed",
}
if err := releaseupdate.WritePendingReport(path, report); err != nil {
t.Fatal(err)
}
g := New("1:test", []int64{42}, slog.Default())
g.UpdateReportPath = path
g.RunningVersions = func() map[string]string { return map[string]string{daemon: installed} }
b, requests := newTelegramTestBot(t)
g.reportPendingUpdate(context.Background(), b)
if len(*requests) != 1 {
t.Fatalf("requests = %#v, want exactly one", *requests)
}
// sendMessage delivers a rich message first, so the operator text is
// the rich_message body (or the plain text field on a fallback).
text := strings.ToLower((*requests)[0].form["rich_message"] + " " + (*requests)[0].form["text"])
for _, want := range test.wantText {
if !strings.Contains(text, want) {
t.Errorf("message = %q, want it to contain %q", text, want)
}
}
for _, deny := range test.denyText {
if strings.Contains(text, deny) {
t.Errorf("message = %q, want it NOT to contain %q", text, deny)
}
}
})
}
}
// TestReportPendingUpdateClearsUnreadableReport: a corrupt report file
// (partial write, disk issue, a future format the running binary doesn't
// understand) must not jam forever -- without clearing it, every future
+41 -13
View File
@@ -598,6 +598,11 @@ type IdentityConfig struct {
type ModelLimits struct {
ContextWindow int `json:"context_window"`
MaxOutputTokens int `json:"max_output_tokens"`
// Reasoning marks a reasoning-class model. It is derived from the model
// catalog (never operator-authored) and carried to the agent worker so a
// task-scoped MCP sampling request omits the max-token bound those models
// reject, exactly as the daemon's chat path does.
Reasoning bool `json:"reasoning,omitempty"`
}
// TaskConfig is the non-secret subset of Config needed to run workflow
@@ -605,17 +610,22 @@ type ModelLimits struct {
type TaskConfig struct {
// BotUser and BotEmail are non-secret commit attribution carried to the
// sandbox because the worker owns local deterministic commits.
BotUser string `json:"bot_user"`
BotEmail string `json:"bot_email"`
Models map[string]string `json:"models"`
ModelLimits map[string]ModelLimits `json:"model_limits,omitempty"`
Budgets Budgets `json:"budgets"`
Dispatch Dispatch `json:"dispatch"`
DiffCapLines int `json:"diff_cap_lines"`
Notify Notify `json:"notify"`
Forge TaskForge `json:"forge"`
ToolPolicy ToolPolicy `json:"tool_policy"`
Review Review `json:"review"`
BotUser string `json:"bot_user"`
BotEmail string `json:"bot_email"`
Models map[string]string `json:"models"`
ModelLimits map[string]ModelLimits `json:"model_limits,omitempty"`
Budgets Budgets `json:"budgets"`
// MaxRetries is the global retry/remediation cap workflow stages read
// (remediate's round cap via Repo.EffectiveMaxRetries). It is carried to the
// worker for the same reason as Budgets: the stage that enforces it runs
// inside archie-agent, not the daemon.
MaxRetries int `json:"max_retries"`
Dispatch Dispatch `json:"dispatch"`
DiffCapLines int `json:"diff_cap_lines"`
Notify Notify `json:"notify"`
Forge TaskForge `json:"forge"`
ToolPolicy ToolPolicy `json:"tool_policy"`
Review Review `json:"review"`
}
// DiffCapOf returns a DiffCapLines value for n. It exists because the field is
@@ -651,6 +661,7 @@ func (c Config) ForTask() TaskConfig {
Models: cloneStringMap(c.Models),
ModelLimits: maps.Clone(c.ModelLimits),
Budgets: c.Budgets,
MaxRetries: c.MaxRetries,
Dispatch: Dispatch{Trigger: c.Dispatch.Trigger, AckReaction: c.Dispatch.AckReaction, Labels: cloneStringMap(c.Dispatch.Labels)},
DiffCapLines: c.DiffCap(),
Notify: c.Notify,
@@ -671,6 +682,7 @@ func (tc TaskConfig) ToConfig() Config {
Models: cloneStringMap(tc.Models),
ModelLimits: maps.Clone(tc.ModelLimits),
Budgets: tc.Budgets,
MaxRetries: tc.MaxRetries,
Dispatch: Dispatch{Trigger: tc.Dispatch.Trigger, AckReaction: tc.Dispatch.AckReaction, Labels: cloneStringMap(tc.Dispatch.Labels)},
DiffCapLines: &tc.DiffCapLines,
Notify: tc.Notify,
@@ -897,6 +909,20 @@ type ContainerConfig struct {
// seeded from this field but validated and stored (and reloaded live)
// separately -- see internal/app/controlplane.
Credentials []CredentialBinding `toml:"credentials" yaml:"credentials" json:"-"`
// RegistryAuth is the credential a private registry requires when the
// configured image (or a profile's) is pulled. It uses the same
// secret-reference shape as every other credential in this file; the
// resolved value is a Docker registry.AuthConfig JSON document, e.g.
// {"username":"bot","password":"..."} ("identitytoken",
// "registrytoken" and "serveraddress" are honoured when present but
// optional -- the daemon derives the host from the image reference). The
// zero value means anonymous pulls, which is what the pool has always
// done; it never fails a deployment that has no private registry.
// RegistryAuth is file-owned, not part of the control-plane JSON document
// (container-runtime-policies): it is a secret reference, resolved at boot
// and restart-required, so the pool receives the resolved value rather
// than a stored one. See internal/container's encodeRegistryAuth.
RegistryAuth SecretRef `toml:"registry_auth" yaml:"registry_auth" json:"-"`
}
// CredentialBinding maps one Kit credential@1 service name to an org secret.
@@ -1141,7 +1167,8 @@ type ChatConfig struct {
// limiting off entirely -- it is an opt-in control, not a default
// throttle that could surprise an existing deployment. This key seeds
// the Channel settings resource, which the Web UI owns at runtime
// (restart-required): editing the file does not change a stored value.
// (live for the channel transports): editing the file does not change a
// stored value.
RateLimit RateLimitConfig `toml:"rate_limit" yaml:"rate_limit"`
}
@@ -1151,7 +1178,8 @@ type ChatConfig struct {
// internal/app/controlplane); the file key is only that resource's seed, and
// the Web UI owns the value a running deployment limits with. Applied by
// boot.startRateLimiter from the config boot.loadRuntimeConfig layers, so a
// stored value takes effect on restart and the resource is restart-required.
// stored change to the limit takes effect on the next Gateway restart; the
// channel transports are the half the Messaging Service reconciles live.
//
// A sender is identified per channel: Telegram by numeric user ID, email by the
// SMTP from address, webhook by the configured route path -- which is a source,
Loaded 100 of 325 files, more files were not shown because too many files have changed in this diff. Show more