Commit Graph
154 Commits
Author SHA1 Message Date
sam f6397af7f4 fix: address the Greptile findings on the merged PRs (#902)
* fix: address the Greptile findings on the merged PRs

Four valid review findings from PRs #892, #893, #898 and #899, which were
merged without the comments being read.

- Config.Clone left the new diff-cap pointers aliased (#899, mine). Clone
  deep-copies every reference field, including the *bool beside it, so a
  clone could write the cap through to the published snapshot. Both the
  top-level field and the identity overrides are copied, with the clone test
  extended to cover them.
- providerOutcomeRecorder.record could report an older call as the latest
  (#892). The completion time is taken before the lock, so two concurrent
  calls can reach the lock in the opposite order to their completion and the
  unconditional store overwrote the newer outcome. It now keeps the newer,
  via a recordAt seam that makes the reordering testable.
- Curator model calls bypassed the recorder (#892). curatorLLMRunner calls
  the runtime directly, so a daemon whose only recent model traffic was
  curator work reported "no calls attempted yet" while the provider was
  demonstrably reachable. It records at its own adapter, and sendChatTurn's
  "single choke point" comment is corrected to say there are two.
- A provider error reached /status raw (#892), unlike channel details, so a
  multiline or oversized error broke the one-line-per-fact layout. Both now
  share one bounded formatter.

Found while fixing the curator path: agentexec.NewRuntime returns nil when no
provider is configured, and a nil *runtime.Runtime panics on its first method
call, so a curator asking for a completion took the daemon down. It now
returns an error, and records nothing -- a local misconfiguration is not a
provider call.

Not fixed, filed instead: the shared lastPollAt masking a stuck identity, the
container max-uptime callback resurrecting teardown bookkeeping, the
dashboard dropping a served-only status filter on cold load, and the route
registry test's line-by-line parser.

* fix(build): drop goimports from the gate's formatter set

`task check` spent 136 of its 167 seconds in `task fmt`, and 134 of those
136 in one formatter: goimports alone over the tree.

golangci-lint's goimports formatter calls imports.Process once per file,
which builds a fresh x/tools ProcessEnv each time, so
internal/imports.getFixesWithSource re-reads modindex.Read($GOMODCACHE) per
file. On this machine that index is 62.7 MiB / 2,081,070 lines (an 11 GiB
module cache), and modindex's in-process memo does not take effect: 3 files
read 187.9 MiB back, 15 files open the payload 15 times. That is ~53 GiB of
I/O and CSV parsing per run, and it is linear at ~155 ms per file (122 files
-> 18.8s, 864 -> 133.8s). Pointing the index cache elsewhere drops the same
command to 3.9s.

The same analyzer runs under `golangci-lint run`, so a cold cache pays it
again: 169s cold with goimports, 31s without -- which is what CI and archied's
own PR gate see, since only `fmt` never caches locally.

Removing it costs no enforcement. gofumpt and gci remain, `task lint` still
reports formatter drift as an issue, and the whole-tree fmt is now 2.2s with
the tree byte-identical. What is given up is goimports' convenience of
adding and removing import lines, which compile errors already cover.

The examples/secret-engines exclusion stays: gofumpt and gci leave those
Yaegi programs alone today, but the exclusion is the stated invariant that a
formatter must not rewrite the imports of files it cannot type-check, and
TestFormattersExcludeInterpretedSecretEngines holds it. Its comment no longer
pins the reason to the formatter that was dropped.

* fix(gateway): stop /status promising and mis-reporting health it cannot see

Two findings from the cloud review of #892.

Title generation no longer records chat-model health. It runs detached after
a first turn under its own 30s bound and its error is swallowed by the
caller, but it shared the process-wide outcome recorder, so a title that
merely timed out overwrote the health line with "failed" -- reporting a
broken provider on the strength of a call no user was waiting for.

The /status command description no longer claims "workers" and "channels".
Neither can render: newStatusHealth reads the container pool and the poll
loop from boot, but the only production router.Health assignment is in
setupGatewayChat, which runs solely in the archie-gateway process, where
neither exists; channels is nil since the Messaging Service took the
front-ends. Only the broker and chat-model lines actually appear, so that is
what the help text now says.

The underlying split -- daemon owns the facts, Gateway owns the Router -- is
filed rather than patched here: it needs either a cross-process health
contract or a ratified two-section /status, not a quick wire-up.
2026-09-20 02:42:36 +10:00
sam c35d9b6fdd feat(messaging): Messaging Service operator surface, chat report rendering, and the diff-cap config fix (#899)
* feat(workflow): post line-anchored review comments with suggested fixes

Bead archie-core-q9au (docs/prds/inline-review.md). A review's findings reached
the operator only as a flat list in the PR body, so the author had to map every
finding back to its line by hand and apply mechanical fixes by hand. Now the
line-anchored findings are also posted as inline review comments, carrying a
one-click suggestion when the reviewer could state the fix.

Domain (internal/domain/workflow):
- ReviewFinding gains Suggestion; Validate rejects one on anything but a
  confirmed finding, so the agent's record_finding call is refused with
  something it can act on. Validate splits into validateAnchor /
  validateClassification / validateDisposition, which also brings it back under
  the cyclop limit the gate enforces.
- StagePostReviewComments runs after StageOpenPR in the implement workflow. It
  posts only line-anchored findings (Line > 0); a whole-file finding has no line
  to attach to and stays in the body list, which remains the complete record.
  It is best-effort: the PR is open and the body already lists every finding, so
  a failed comment is logged, never a park -- the same reasoning as OpenPR's
  existing best-effort LinkBranch call.
- The suggestion fence is gated on Verdict == confirmed where the fence is
  written, because it is a one-click apply: a wrong suggestion breaks the code
  and the author has to notice.

Forge (internal/forge): ReviewCommentWriter + GitHub/Gitea implementations,
type-asserted like PullRequestReader so a forge that cannot is skipped rather
than faked (NoopForge is pinned as not satisfying it). GitHub loops one
CreateComment per comment and joins the failures, so one refused anchor cannot
drop the rest; Gitea carries them on a single COMMENT-state review.

Two deliberate departures from the PRD's sketch, both recorded in it:
- The interface is batch (CreateReviewComments), not one call per comment:
  Gitea's only inline shape is one submitted review, and the set then crosses
  NATS once instead of once per finding.
- The head SHA is resolved inside each forge implementation rather than by the
  caller. The stage runs in archie-agent, which holds no forge credentials and
  has no PullRequestReader; the daemon-side forgerpc handler is the only place
  that does.

Transport (internal/forgerpc): workflow.Forger gains the method, so the agent
reaches the daemon over the new create_review_comments subject; the server
type-asserts the writer and answers an incapable forge with an error the stage
logs. This is the one proxied method whose absence on an image skew does not
park a task.

The reviewer prompt and record_finding schema ask for a suggestion on a
confirmed, mechanically fixable defect only.

Regenerating the wfextract symbol table -- the documented way a new exported
workflow symbol reaches interpreted stage plugins -- also picked up symbols that
were missing from the committed table (the Status* and Source* constants,
ReviewCheck, Store, Task, and the _Store wrapper). Nil-guards were re-applied to
every wrapper afterwards, as that file's header requires.

Evidence: task check exit 0 (incl. 229 UI tests). Three mutations of the new
stage logic each fail a test that names the behaviour -- dropping the
confirmed-only gate, dropping the Line > 0 filter, and letting a posting failure
escape -- so the tests discriminate rather than passing vacuously.

* fix(build): drop the Dockerfile pin assertion

The guard fetched golangci-lint's go.mod at image build time to fail the build
when GOLANGCI_LINT_VERSION and GOFUMPT_VERSION disagreed. It only ever tripped
on self-inflicted pin drift, but it added a network fetch and a new failure
mode to the release build, which is not where that belongs.

The pins stay correct (golangci-lint 2.13.2, whose vendored gofumpt v0.11.0
equals GOFUMPT_VERSION) and the comment above them still states the invariant,
so the reasoning survives without the build-time machinery. The pin change
itself reaches the agent image through the normal release path: CI rebuilds
and pushes archie-agent on an archie/vX.Y.Z release, and a gateway-only
release deliberately does not.

* feat(workflow): add the remediate workflow for PR review remediation

Implements step 5 of docs/prds/pr-review-remediation.md: a remediate
workflow that resumes the task's existing worktree onto its PR branch
(worktree.Manager.Resume, not Prepare's refresh), runs a builder agent
against the review's actionable comments, commits and pushes to the
same branch, and replies via the forge's ReplyToReview. Task.RetryCount
doubles as the remediation round counter, capped by the repo's
max_retries; the round-cap stage parks the task and posts one comment
when it's hit.

Task gains a ReviewPayload field (JSON-encoded review unit) with full
store/proto/staterpc plumbing. workflow.Forger gains Comment and
ReplyToReview, proxied to archie-agent via a new forgerpc RPC. Resume
and its Dir lookup are added as an unexported optional Trees capability
rather than widening the exported interface, which is already at
golangci-lint's interfacebloat cap and is projected into wfextract's
Yaegi symbol table for repository-authored stage plugins.

The daemon-side reaction consumer that dispatches into this workflow
(subscribing ARCHIE_REACTIONS, grouping comments by review, enforcing
per-task serialization) is tracked separately in archie-core-8li9.4.

* feat(messaging): extract standalone messaging service binary and process composition

* feat(messaging): adapt channel frontends to consume Gateway ChatContract

* feat(messaging): remove in-process channels and Router from daemon bootstrap

Telegram, email and webhook now run only in the Messaging Service, which
reaches this process through the Gateway's ChatContract. The daemon builds
no channel and no gateway.Router, so the wiring that existed only to serve
them goes with it: setupGateways, setupEmailGateway, setupWebhookGateway,
setupTelegramGateway, buildTelegramRouter, makeTelegramReload and the
release-announcement path helper.

The seams those paths owned move to where they can still be honoured:

- Channel ValidateConfig runs in archiemessaging.compose, which now returns
  an error, so a configured-but-invalid front-end fails the service rather
  than being dropped silently.
- The inbound rate limiter moves to the Gateway process, which owns the only
  Router and therefore the only place a per-sender budget applies to every
  channel's turns.
- The daemon's gateway readiness probe becomes a ContractProbe against the
  Gateway. It cannot observe a channel lifecycle it no longer runs, and a
  channel-count probe would have reported every daemon degraded.
- telegramSetup is now chatSetup (in chat_setup.go), trimmed to the fields a
  chat turn runner actually reads.

Telegram operator features that were wired only in the daemon (/update,
release announcements, version reporting, SIGHUP reload) are not yet wired
in the Messaging Service: archie-core-8cda.6.6. Channel lifecycle reporting:
archie-core-8cda.6.8. The Gateway process builds no worktree manager, so
review_pr is unadvertised on every channel: archie-core-8cda.6.7.

* feat(messaging): re-wire the Telegram operator surface in the Messaging Service

setupTelegramGateway satisfied eight telegram.Gateway seams from the daemon's
own process. Each is resolved by where the fact it needs is owned, not by
where it used to be built; the decision table is in
docs/architecture/migration-decisions.md.

Wired here: /update (built from this service's own [chat.telegram] check and
install commands), the update-report relay (same sha256(bot_user)[:8] path the
daemon wrote, so a pending report is found rather than written afresh),
show_tool_calls, and /restart's config reload, which re-resolves the token and
allowlist from this service's own file and refuses a reload that resolves to
an empty token.

/version is sourced from the Gateway: ChatSnapshot gains a Version field
carrying the Gateway's rendered build block, read per call and bounded by the
dependency timeout. cmd/archie-messaging deliberately takes no version
ldflags, so it cannot report a partial upgrade as a matched one.

RunningVersions and ReleaseAnnouncements stay nil. Both need a component's
real running version, and the only versions this process can obtain are a
different binary's stamps -- reporting them would manufacture exactly the
false success the update check exists to catch, and the announcer without
parseable versions is a no-op that reads as configured. archie-core-8cda.6.9
carries the contract that unblocks them. Dangerous stays nil: no composition
has ever supplied a command authority.

Enrich is dropped from the update service: componentInstallTypeEnricher read
[nats], which this service must not decode, so the check command's own
reported install type stands.

* fix(workflow): guard inline review comments against head drift

Two findings from the review of q9au.

A comment's line numbers describe exactly one revision of the branch, but the
forge resolved the pull request's head at posting time -- after the review had
already read the worktree. A push in that window (a collaborator, a retry) meant
the findings were anchored to whatever now occupied those lines, mislabelling
instead of failing. The revision the review measured now travels with the
comment set: OpenPR records it from the same worktree read it already performs
(TaskContext.ReviewedHeadSHA), StagePostReviewComments threads it through
Forger to forgerpc, and the forge refuses the whole set once the head it reads
has moved off it -- which the best-effort stage logs, the PR body remaining the
record. An unmeasured revision posts unverified rather than dropping every
finding, so a worktree that cannot report one degrades instead of regressing.

Suggestion was documented single-line only, because GitHub applies a suggestion
fence up to its first line and a multi-line replacement would silently corrupt
the file. Only the verdict was validated, so a multi-line suggestion reached the
renderer and posted as that truncated fence. Validate now rejects an embedded
newline, which makes record_finding refuse it with feedback the model can act on
instead of the reviewer discovering the corruption later.

* fix(gateway): render chat reports as lines, not one run-on paragraph

/status and /tasks are line-oriented reports, but their newlines were plain
"\n" -- CommonMark's soft wrap. The Telegram rich-block renderer space-joins
soft-wrapped lines, by design: a Telegram paragraph renders an embedded
newline as nothing at all, so a space is the only separator that survives.
The reports therefore arrived as one line each: "Runtime Provider: DeepSeek
Model: deepseek/deepseek-flash", and every task's title, state and park
reason jammed together.

The renderer now implements CommonMark's hard line break (two trailing
spaces, or a trailing backslash): the line closes its own block and the next
one hugs it, so a stanza keeps its lines without the spacer paragraph that
separates unrelated blocks. Ordinary prose is untouched -- a model's wrapped
paragraph still joins with a space. The producers mark their report lines
with it.

Two report fixes alongside:

- A park reason is capped to its first line at 160 runes. It carries whatever
  the failing stage reported, which for a lint gate is the whole tool log;
  unbounded, one parked task pushed every other task out of the reply. The
  full reason is still on the task.
- /status loses its "Runtime" sub-header. A section header over two fields
  made a four-fact report read like a document.

* fix(config): let diff_cap_lines = 0 actually switch the cap off

Config.DiffCapLines was a plain int, so the defaults pass could not tell an
absent key from an explicit 0 and rewrote both to 400. The dashboard schema
said "0 means unlimited" and StageDiffCap treated <= 0 as unlimited, but 0
never reached the step: the documented way to switch the cap off was
unreachable through config.

Same root cause, worse instance: configForIdentity assigned
root.DiffCapLines = identity.DiffCapLines unconditionally, so an identity
that simply omitted the key overwrote the shared cap with the zero value and
silently disabled the safety rail for that identity.

DiffCapLines is now *int on both Config and IdentityConfig -- nil is absent,
0 is no cap -- read through Config.DiffCap(). Defaults fill only nil, and an
identity overrides only when it set the key. TaskConfig keeps a resolved int,
so the daemon/agent wire format is unchanged and a version-skewed agent
cannot misread a null.

The park reason changes too. It told the operator to "approve manually", but
a parked task's actions are retry, abandon and reject (internal/taskstate):
there is no approve path out of diff-cap, so it sent them looking for a
button that does not exist. It now names the two things that work.

* feat(config): make the daemon's readiness dependency timeout configurable

The daemon's gateway probe bounded its call with a hardcoded 5s constant,
while every other NewContractProbe call site (archie-ui, archie-messaging)
passes the service's configured DependencyTimeout. It was the only magic
number among them.

[health].dependency_timeout carries it, defaulting to the same 5s the
extracted services' -dependency-timeout flag defaults to, so every process
bounds a probe into a dependency the same way out of the box.

It is not added to the dashboard config schema or the reloadable set:
[health] appears in neither today, health.listen included, and a probe
assembled at boot would not observe a reload anyway.
2026-09-20 00:32:13 +10:00
sam 800352a692 feat(gateway): give /status a real daemon-health surface (#892)
* feat(gateway): give /status a real daemon-health surface

Phase 2 of archie-core-wp9s. Phase 1 split the commands apart (health vs
work) but left /status with one aggregate queue line and a static runtime
block; it could not answer the questions an operator actually asks when
something feels wrong. /status now reports, between the queue line and the
runtime block:

- Broker: the live state of this process's own NATS connection
  (eventbus/nats.Client.Connected), not a fresh dial -- a probe would
  report the probe's reachability, not the daemon's.
- Containers: the pool's own active counter against the configured cap
  (Pool.Active/Cap), not a Docker listing, which would include containers
  this pool does not own. A zero cap means unlimited, so it is not shown
  as a ratio.
- Channels: the channel manager's real lifecycle state, including a
  failed adapter's reason (the documented decision: read the existing
  manager rather than re-deriving health per adapter).
- Chat model: last-known outcome of every chat-model call this process
  made, recorded once in sendChatTurn (the single choke point) and read
  back by /status. The second documented decision: never a live probe,
  so /status stays instant and reports what happened -- or says plainly
  that nothing has been attempted yet.
- Last poll: when the most recent poll pass began, stamped on both poll
  paths (Run's single-identity loop and runIdentities' per-identity
  goroutines). A pass that hangs leaves the stamp stale, which is the
  signal; stamping completion would leave a wedged poller looking
  healthy.

Truthful or absent: a section this process has no source for is omitted
rather than rendered as a zero. The standalone Gateway owns no poll loop,
pool or channel manager, so its /status reports broker and chat-model
health only; aggregating across processes would need a new daemon RPC and
is deliberately not built. Version stays out (/version owns it).

/status and /tasks descriptions now describe what each command actually
does on both published surfaces (Telegram menu/help and the shared
specs), pinned by a test that also fails if the two surfaces drift.

Sources are read through functions, not captured values: occupancy and
channel state move, and the daemon is built after the gateways, so a
source captured at construction would freeze /status on boot state.

Left in place deliberately: the Runtime (provider/model) block, which
Phase 1 flagged and which duplicates /model and /whoami. Removing it is a
product call on a command Sam uses daily, not a mechanical follow-up.

* docs(prds): record the two open /status health decisions as settled

Both were flagged in PR #892 as needing the maintainer's call; both are now
answered, and the PRD is the place a follow-up would look:

- The Runtime (provider/model) block stays. It is the one line in /status the
  ticket's own rule says should not be there, but /status is read daily to see
  the active model at a glance, so the duplication is deliberate. Recorded as a
  settled product decision so it is not removed later as an oversight.
- Health stays per-process. Pooling the daemon's and the Gateway's chat-model
  outcomes would need a new daemon<->gateway health RPC for one status line, and
  a per-process line is true where an aggregate would be a claim about a process
  the answering one cannot see.

/agents' published description is still left to archie-core-mxls, unchanged.
2026-09-20 00:30:23 +10:00
sam e0b6aa6105 feat(workflow): post line-anchored review comments with suggested fixes (#891)
* feat(workflow): post line-anchored review comments with suggested fixes

Bead archie-core-q9au (docs/prds/inline-review.md). A review's findings reached
the operator only as a flat list in the PR body, so the author had to map every
finding back to its line by hand and apply mechanical fixes by hand. Now the
line-anchored findings are also posted as inline review comments, carrying a
one-click suggestion when the reviewer could state the fix.

Domain (internal/domain/workflow):
- ReviewFinding gains Suggestion; Validate rejects one on anything but a
  confirmed finding, so the agent's record_finding call is refused with
  something it can act on. Validate splits into validateAnchor /
  validateClassification / validateDisposition, which also brings it back under
  the cyclop limit the gate enforces.
- StagePostReviewComments runs after StageOpenPR in the implement workflow. It
  posts only line-anchored findings (Line > 0); a whole-file finding has no line
  to attach to and stays in the body list, which remains the complete record.
  It is best-effort: the PR is open and the body already lists every finding, so
  a failed comment is logged, never a park -- the same reasoning as OpenPR's
  existing best-effort LinkBranch call.
- The suggestion fence is gated on Verdict == confirmed where the fence is
  written, because it is a one-click apply: a wrong suggestion breaks the code
  and the author has to notice.

Forge (internal/forge): ReviewCommentWriter + GitHub/Gitea implementations,
type-asserted like PullRequestReader so a forge that cannot is skipped rather
than faked (NoopForge is pinned as not satisfying it). GitHub loops one
CreateComment per comment and joins the failures, so one refused anchor cannot
drop the rest; Gitea carries them on a single COMMENT-state review.

Two deliberate departures from the PRD's sketch, both recorded in it:
- The interface is batch (CreateReviewComments), not one call per comment:
  Gitea's only inline shape is one submitted review, and the set then crosses
  NATS once instead of once per finding.
- The head SHA is resolved inside each forge implementation rather than by the
  caller. The stage runs in archie-agent, which holds no forge credentials and
  has no PullRequestReader; the daemon-side forgerpc handler is the only place
  that does.

Transport (internal/forgerpc): workflow.Forger gains the method, so the agent
reaches the daemon over the new create_review_comments subject; the server
type-asserts the writer and answers an incapable forge with an error the stage
logs. This is the one proxied method whose absence on an image skew does not
park a task.

The reviewer prompt and record_finding schema ask for a suggestion on a
confirmed, mechanically fixable defect only.

Regenerating the wfextract symbol table -- the documented way a new exported
workflow symbol reaches interpreted stage plugins -- also picked up symbols that
were missing from the committed table (the Status* and Source* constants,
ReviewCheck, Store, Task, and the _Store wrapper). Nil-guards were re-applied to
every wrapper afterwards, as that file's header requires.

Evidence: task check exit 0 (incl. 229 UI tests). Three mutations of the new
stage logic each fail a test that names the behaviour -- dropping the
confirmed-only gate, dropping the Line > 0 filter, and letting a posting failure
escape -- so the tests discriminate rather than passing vacuously.

* fix(build): drop the Dockerfile pin assertion

The guard fetched golangci-lint's go.mod at image build time to fail the build
when GOLANGCI_LINT_VERSION and GOFUMPT_VERSION disagreed. It only ever tripped
on self-inflicted pin drift, but it added a network fetch and a new failure
mode to the release build, which is not where that belongs.

The pins stay correct (golangci-lint 2.13.2, whose vendored gofumpt v0.11.0
equals GOFUMPT_VERSION) and the comment above them still states the invariant,
so the reasoning survives without the build-time machinery. The pin change
itself reaches the agent image through the normal release path: CI rebuilds
and pushes archie-agent on an archie/vX.Y.Z release, and a gateway-only
release deliberately does not.

* feat(workflow): add the remediate workflow for PR review remediation

Implements step 5 of docs/prds/pr-review-remediation.md: a remediate
workflow that resumes the task's existing worktree onto its PR branch
(worktree.Manager.Resume, not Prepare's refresh), runs a builder agent
against the review's actionable comments, commits and pushes to the
same branch, and replies via the forge's ReplyToReview. Task.RetryCount
doubles as the remediation round counter, capped by the repo's
max_retries; the round-cap stage parks the task and posts one comment
when it's hit.

Task gains a ReviewPayload field (JSON-encoded review unit) with full
store/proto/staterpc plumbing. workflow.Forger gains Comment and
ReplyToReview, proxied to archie-agent via a new forgerpc RPC. Resume
and its Dir lookup are added as an unexported optional Trees capability
rather than widening the exported interface, which is already at
golangci-lint's interfacebloat cap and is projected into wfextract's
Yaegi symbol table for repository-authored stage plugins.

The daemon-side reaction consumer that dispatches into this workflow
(subscribing ARCHIE_REACTIONS, grouping comments by review, enforcing
per-task serialization) is tracked separately in archie-core-8li9.4.

* feat(messaging): extract standalone messaging service binary and process composition

* feat(messaging): adapt channel frontends to consume Gateway ChatContract

* feat(messaging): remove in-process channels and Router from daemon bootstrap

Telegram, email and webhook now run only in the Messaging Service, which
reaches this process through the Gateway's ChatContract. The daemon builds
no channel and no gateway.Router, so the wiring that existed only to serve
them goes with it: setupGateways, setupEmailGateway, setupWebhookGateway,
setupTelegramGateway, buildTelegramRouter, makeTelegramReload and the
release-announcement path helper.

The seams those paths owned move to where they can still be honoured:

- Channel ValidateConfig runs in archiemessaging.compose, which now returns
  an error, so a configured-but-invalid front-end fails the service rather
  than being dropped silently.
- The inbound rate limiter moves to the Gateway process, which owns the only
  Router and therefore the only place a per-sender budget applies to every
  channel's turns.
- The daemon's gateway readiness probe becomes a ContractProbe against the
  Gateway. It cannot observe a channel lifecycle it no longer runs, and a
  channel-count probe would have reported every daemon degraded.
- telegramSetup is now chatSetup (in chat_setup.go), trimmed to the fields a
  chat turn runner actually reads.

Telegram operator features that were wired only in the daemon (/update,
release announcements, version reporting, SIGHUP reload) are not yet wired
in the Messaging Service: archie-core-8cda.6.6. Channel lifecycle reporting:
archie-core-8cda.6.8. The Gateway process builds no worktree manager, so
review_pr is unadvertised on every channel: archie-core-8cda.6.7.

* feat(messaging): re-wire the Telegram operator surface in the Messaging Service

setupTelegramGateway satisfied eight telegram.Gateway seams from the daemon's
own process. Each is resolved by where the fact it needs is owned, not by
where it used to be built; the decision table is in
docs/architecture/migration-decisions.md.

Wired here: /update (built from this service's own [chat.telegram] check and
install commands), the update-report relay (same sha256(bot_user)[:8] path the
daemon wrote, so a pending report is found rather than written afresh),
show_tool_calls, and /restart's config reload, which re-resolves the token and
allowlist from this service's own file and refuses a reload that resolves to
an empty token.

/version is sourced from the Gateway: ChatSnapshot gains a Version field
carrying the Gateway's rendered build block, read per call and bounded by the
dependency timeout. cmd/archie-messaging deliberately takes no version
ldflags, so it cannot report a partial upgrade as a matched one.

RunningVersions and ReleaseAnnouncements stay nil. Both need a component's
real running version, and the only versions this process can obtain are a
different binary's stamps -- reporting them would manufacture exactly the
false success the update check exists to catch, and the announcer without
parseable versions is a no-op that reads as configured. archie-core-8cda.6.9
carries the contract that unblocks them. Dangerous stays nil: no composition
has ever supplied a command authority.

Enrich is dropped from the update service: componentInstallTypeEnricher read
[nats], which this service must not decode, so the check command's own
reported install type stands.

* fix(workflow): guard inline review comments against head drift

Two findings from the review of q9au.

A comment's line numbers describe exactly one revision of the branch, but the
forge resolved the pull request's head at posting time -- after the review had
already read the worktree. A push in that window (a collaborator, a retry) meant
the findings were anchored to whatever now occupied those lines, mislabelling
instead of failing. The revision the review measured now travels with the
comment set: OpenPR records it from the same worktree read it already performs
(TaskContext.ReviewedHeadSHA), StagePostReviewComments threads it through
Forger to forgerpc, and the forge refuses the whole set once the head it reads
has moved off it -- which the best-effort stage logs, the PR body remaining the
record. An unmeasured revision posts unverified rather than dropping every
finding, so a worktree that cannot report one degrades instead of regressing.

Suggestion was documented single-line only, because GitHub applies a suggestion
fence up to its first line and a multi-line replacement would silently corrupt
the file. Only the verdict was validated, so a multi-line suggestion reached the
renderer and posted as that truncated fence. Validate now rejects an embedded
newline, which makes record_finding refuse it with feedback the model can act on
instead of the reviewer discovering the corruption later.
2026-09-19 23:37:23 +10:00
sam d2504d8023 chore(plugin): delete the unwired capability-host health surface (#882)
* chore(plugin): delete the unwired capability-host health surface

plugin.Host.Health, plugin.Host.Manifests, ModuleStatus and their private
safeHealth helper were implemented, exhaustively tested, and called from nowhere
but their own tests. AGENTS.md forbids speculative surfaces, and no settled
design places plugin module status anywhere that exists:

- docs/prds/status-health-surface.md Phase 2 is self-described as "a proposed
  resolution, awaiting Sam's sign-off -- not yet implemented", and does not
  mention plugin status even in that unapproved list.
- The readiness subsystem enumerates exactly five probes -- state_db, config,
  disk, model, gateway (internal/app/archied/readiness.go) -- with no plugin
  probe.
- docs/architecture/plugins-and-extensions.md names no operator-visible plugin
  status surface.
- The only bead naming it is closed.

plugin.Host itself stays: it is the daemon's live capability host, constructed
and Start/Stop-ed from internal/app/archied. Only the unconsumed observability
methods go.

host_contract_test.go encoded the defect -- it asserted plugin.Host *defines*
"Health" and "Manifests" -- so those two names are removed from its list, which
is now exactly the shipped lifecycle: Register, Start, Stop. The three tests
covering the deleted surface are removed with it.

The Yaegi symbol table (internal/plugin/pluginextract) was regenerated rather
than hand-edited, so ModuleStatus leaves the reflected type set with it.

Reachability evidence: Host.Health and Host.Manifests appear only in their own
tests; plugin.Host is otherwise reached from bootstrap.go and gateway.go for
Register/Start/Stop. Note that "task reachability" does NOT flag these -- the
address-taken constructor is a known blind spot of the whole-program RTA.

* chore(plugin): delete the dead Module.Health chain wired to nothing

Extends the capability-host observability deletion: plugin.Module.Health
was a required interface method with real implementations in
tools/provider/mcp and tools/provider/builtin, aggregated by
tools/provider.Registry.Health via a provider-package-local safeHealth,
and backed by per-module LifecycleState bookkeeping in plugin.Host
(registeredModule.state, setModuleState). None of it had a caller before
this change -- Host.Health was the only thing that would ever have read
a registered module's Health(), and Host.Health was already deleted as
unreachable. registeredModule.state was write-only: nothing ever read it.

Deleted:
- plugin.Module.Health, plugin.Health/HealthStatus (+ consts),
  plugin.LifecycleState (+ consts), legacyModule.Health,
  registeredModule.state, Host.setModuleState and its call sites.
- tools/provider.Registry.Health, tools/provider.safeHealth.
- tools/provider/mcp.Provider.Health, tools/provider/builtin.Provider.Health.
- Tests exclusively covering the above, and fakes' now-unnecessary Health
  methods (plugin/host_test.go, pluginextract/reachability_test.go,
  tools/provider/{provider,optional}_test.go, tools/provider/mcp/provider_test.go,
  app/agentworker/mcp_providers_test.go).

internal/plugin/pluginextract's generated file was regenerated via
`go generate ./internal/plugin/...` and the yaegi wrapper nil-guards
(dropped by fresh generation, as documented in reachability_test.go)
were reapplied by hand to WManifest/WStart/WStop/WName/WVersion,
following this repo's established post-regen pattern.
2026-09-19 14:10:39 +10:00
sam 04d4d8f1d8 chore(memory): delete internal/memory (unification slice 5) (#889)
Deletes internal/memory (Manager, builtin.Provider, memory_edit tool) now
that slices 2-4 built the CRUD engine family and its per-turn read/write
seams. Also deletes internal/infrastructure/memory/builtin/provider.go (the
memory_edit adapter over the old package -- its markdown store already
relocated in slice 2), internal/tools/provider/memory (the old tool
registration, redundant with slice 4's memory_create/update/delete/list),
config.Memory.{Provider,ProviderConfig} and the validateMemory rejection
that existed only to complain about Provider, and webui.MemoryStatus/
MemoryProviderHandle plus its dashboard page -- never wired in production
(s.Memory was always nil), confirmed by /api/capabilities always reporting
"memory": false.

Rewrites ARCHITECTURE.md's "Memory engine family" section to describe the
current engine (internal/domain/memory contract, internal/infrastructure/
memory builtin engine + its builtin/ markdown store, the per-turn read seam
in gateway/turn_memory.go, the per-turn write tools in
gateway/turn_memory_tool.go) instead of the deleted Manager. Fixes two
docs/architecture/configuration.md rows that described the deleted
manager's literals and SaveConfig mechanism.
2026-09-19 13:43:16 +10:00
sam a18067d85d feat(gateway): chat memory write path (unification slice 4) (#888)
Build a per-turn memory tool set (memory_create/update/delete/list) onto
the turn's resolved Subject, matching slice 3's read-side scope model.
The model picks a scope KIND from an enum built off Subject.WritableScopes()
(never global); the agent and user ids a scope resolves to come from the
turn's Subject, with no schema field either could ever land in. A subject
with no writable scope, or a nil write store, gets no memory tools at all.

TurnRunnerConfig gains MemoryWriter (MemoryWriteStore: Create/Update/
Forget/List), wired alongside the existing MemoryEngine read surface
through the same active engine (boot.activeMemoryEngine, narrowed by
memoryStore/memoryWriter).
2026-09-19 12:59:11 +10:00
sam cc7c23a58c feat(gateway): chat memory read path (unification slice 3) (#887)
Wires the durable-memory engine into the chat turn: prepareTurn resolves a
per-channel Subject (BotUser as AgentID, TurnRunnerConfig.UserIdentity for
the user), runs a scope-only Query over Subject.Scopes(), and renders the
result into a new <memory purpose="durable_context" trust="data"> block
between <tools> and <env>, following the same untrusted-data framing as
those two blocks. The read is synchronous, total, and never surfaces an
error: a nil engine, a failing Query, or a panicking provider all degrade to
no memory block (recovered in gateway.renderMemory).

UserIdentity is supplied per channel by the composition
(userIdentityResolver in internal/app/archied): telegram and email resolve
their native SenderID, the dashboard ("web") and webhook resolve nothing, so
a webhook's route-path SenderID can never become a user identity and the
dashboard's shared bearer token can never impersonate a person.

Fixes the composition-ordering bug that left every turn runner holding a nil
memory engine: setupMemoryAll now runs before setupGateways in main.go, and
before setupGatewayChat in gateway.go's startGatewayRuntime, so
boot.memoryStore() has a live engine at every turn-runner construction site.

Named constants bound the read: memoryRecordsPerScope per scope, and
memoryBlockByteCap on the rendered block, which drops (and logs) trailing
records rather than truncating one mid-line.
2026-09-19 11:40:45 +10:00
sam 7ee5025d3c feat(memory): four-scope contract and builtin engine CRUD (unification slice 2) (#886)
* feat(memory): four-scope contract with CRUD and revisions

Seam for memory unification slice 2 (docs/prds/memory-engine-unification.md
§1, §2, §7): internal/domain/memory is now a scope-addressed CRUD contract.

- Scope{Kind,Agent,User} with the four ratified kinds, Validate rejecting a
  component present where it is not required, and a length-prefixed Key so
  opaque channel-native ids cannot collide.
- Subject{AgentID,UserID} with Scopes()/WritableScopes(): global is never
  writable, and an absent user id removes every scope requiring one.
- Store becomes Create/Get/Query/List/Update/Forget/Revisions over
  NewRecord/RecordUpdate/Record/Revision, with ErrNotFound/ErrStaleRevision.
  Observation, Identity and Metadata are deleted.
- The content scanner moves into the family (internal/domain/memory) so every
  engine applies it in Create and Update; internal/memory re-exports the types
  until slice 5 deletes it.
- The markdown store moves to internal/infrastructure/memory/builtin, its
  correct home per organisation.md.

WIP: internal/infrastructure/memory and internal/infrastructure/sessioncurator
are converted in the lanes that follow this commit.

* feat(memory): builtin engine CRUD, retained revisions and the curator's agent-user write

Slice 2 implementation (docs/prds/memory-engine-unification.md §1, §2, §6,
§7), on top of the contract seam in the previous commit.

- internal/infrastructure/memory: per-scope markdown heads with a JSON
  provenance marker, per-scope HISTORY.md retaining superseded and deleted
  states, Update superseding before replacing the live block, ErrStaleRevision
  via RecordUpdate.Expected, idempotent Forget, and the family scanner applied
  in Create and Update. Scope directories are hex(sha256(Scope.Key())).
- internal/infrastructure/sessioncurator: the participant is derived from the
  session's user-role senders; exactly one writes agent-user scope, zero or
  several write nothing and return ActionSkipped naming the count.
- docs/architecture: the settled decision recorded in migration-decisions §5
  (superseding the 2026-09-02 marker entry) and agent-system.md pointing at it.

Draft: to be squashed and reviewed before publication.

* test(curator): conform the registry's fake memory engine to the CRUD contract

The fake still implemented the pre-CRUD Store surface (Write taking an
Observation, List/Forget keyed by a bare string), so the package's test
binary no longer compiles: fakeMemoryEngine is missing Create, Get, Update
and Revisions, and domainmemory.Observation no longer exists. That took
`go fix ./...`, `go vet ./...` and `go test ./...` down with it, which is
what `task check` runs first.

Missed by the same commit that updated internal/infrastructure/sessioncurator's
own fake to the new interface. Repaired on the fake, not by loosening the
interface: the fake is the only thing out of date.

* fix(sessioncurator): refuse to attribute a session with unidentified user messages

sessionParticipant skipped user-role messages with an empty SenderID and
counted only the identified ones, so a session holding one identified sender
plus any number of unidentified user messages yielded that sender as the
single participant. Those messages are still part of the excerpt the
extraction model reads, so facts derived from content of unknown provenance
were written into the identified sender's agent-user scope: user A's memory
can end up crediting content that may belong to someone else. It is exactly
the case the function's own doc comment claims to handle, which it only
caught when no message at all was identified.

An unidentified user message now makes the session unattributable outright,
beside any number of identified senders, and the skip action's reason names
the count instead of reporting a sender count that no longer explains the
skip. Assistant-role messages are unaffected: they carry the bot, so they
still cannot manufacture a participant.

Fixed in sessionParticipant rather than at the write, because it is the
single producer of the attribution decision: it now returns the participant
or the reason there is none, so the caller cannot print a reason that
disagrees with what it decided. The fail-closed path (nothing written, one
ActionSkipped) and the extraction flow are unchanged.

* fix(memory): carry record content through the block format intact

builtin.Store's format ends a block at a blank line and starts a section at a
"## " line, but storeContent only trimmed the content and rejected an empty
one. Content holding either delimiter was accepted, written, and answered back
to the caller in full -- then read back truncated at the delimiter by Get,
List, or a process restart, with the remainder unaddressable and no error
anywhere along the path.

Fixed at the producer, in the render/parse pair that owns the block format:
renderBlock escapes the two delimiter shapes (an empty line becomes a single
backslash, a line already starting with a backslash or "## " gains one) and
parseBlocks undoes it on the way back, so a record's content survives the
write whole. Rejecting the content instead would have made an ordinary
multi-paragraph note or a quoted heading unrepresentable; escaping every line
would have changed the on-disk shape of every record. Only the lines that need
it are escaped, so a one-paragraph record is byte-identical to what the
unfixed engine wrote and the document stays hand-editable.

Reachable from both producers: the chat tool's note and the curator's
model-extracted fact are free text, and neither is constrained to one
paragraph.

Tests: TestBuiltinEngineRenderedBlockRoundTripsContent asserts
parseBlocks(renderBlock(c)) == c for eleven content shapes, and
TestBuiltinEngineContentSurvivesTheWriteAndAReopen carries the same shapes
through Create, Get, and a second engine reading the file a restart would.

* fix(memory): compare Update's expected revision against the file, not the cache

Update read the record to compare against out of findBlock(stores.live, ...)
-- the engine's cached copy of the scope document -- and checked Expected
against that copy. This engine caches one document per scope and rewrites it
whole, so a second engine over the same scope file (archied and the gateway,
or a restart) holds a copy taken when it last read the file: its view of the
record's revision is superseded, Expected matches the revision it remembers,
ErrStaleRevision never fires, and the rewrite that follows writes that stale
copy back over the file -- discarding the other writer's update and every
record written since the copy was loaded. The field existed and could not
fire.

Reload the scope document before anything is decided from it, so the check
answers about the revision actually on disk and the rewrite starts from the
current document. That is the smaller of the two fixes the PRD leaves open
("Not determined": a reload-before-mutate or an advisory lock); no locking
scheme, no retry loop, and no change to the read paths.

Still unsolved, and deliberately: the whole-document rewrite is
last-writer-wins between the reload and the write, so two engines writing one
scope concurrently can still lose one of the two writes, and HISTORY.md is
rewritten from the cached copy the same way.

Tests: TestBuiltinEngineSecondEngineCannotReuseAStaleRevision asserts the
second engine gets ErrStaleRevision and the first writer's state survives;
TestBuiltinEngineUpdateDoesNotEraseRecordsItHasNotLoaded asserts the record
written after the second engine loaded is still there after its update.

* fix(memory): emit the scanner's warning instead of dropping it

scanContent returned nil for every result that was not ThreatBlock, so a
ThreatWarn result -- documented in the scanner as "allow the write but emit a
warning", and what every sensitive-data match produces (API keys, passwords,
tokens, private keys) -- was stored with no audit signal anywhere. The
comment justified the silence by pointing at Bind, but Bind's argument is
about Registrar.Events: internal/events declares no memory kind and the
composition binds no memory event sink, so an invented kind would look wired
and not be. A scanner warning is a diagnostic rather than an event, and there
was nothing else to carry it: the registrar had no logger.

Bind now takes Registrar.Log, optional and mirroring the curator family's
Registrar for the same reason, scanContent emits the warn-level result
through it before the write is allowed, and bootstrap passes the daemon's
logger so the line reaches archied's configured sinks instead of only the
process default. The write is still allowed: only the silence changed.

Tests: TestBuiltinEngineScanWarnsWithoutRefusingTheWrite runs Create and
Update over warn, clean and block content, asserting the write lands either
way, that a warn-level scan result logs exactly one warning carrying the
scanner's own message, that clean content logs none, and that a refused
block-level match logs none.

* test(memory): carry every content shape through Update and Revisions

Update replaces the block it read out of the document by that text, and
retains the superseded state in HISTORY.md, so an escaped block has to be
found, superseded and retained exactly like an unescaped one. Covered here
rather than left to the Create path: without the escaping, every shape in
this table came back truncated after an update, including the plain
one-paragraph record whose replacement holds a blank line.

The test asserts values only: the updated record as a second engine reads it
off disk, and the superseded content as Revisions returns it.

* style(memory): apply the linter's gofumpt form to the Update revision error

golangci-lint's bundled gofumpt wants the closing paren of that wrapped
fmt.Errorf on its own line; the standalone gofumpt binary the fmt task uses
(0.7.0, a different version) leaves it as it was. The text is unchanged from
before the engine fixes in this branch, so this is pre-existing drift inside a
function these commits already touch -- taken so the file is clean under both
formatters rather than only one.
2026-09-19 11:13:21 +10:00
sam b332f03175 Merge pull request #878 from samcharles93/feat/attempt-attribution
Feat/attempt attribution
2026-09-19 02:48:25 +10:00
sam 8d3925a844 Merge pull request #873 from samcharles93/fix/restart-command-wiring
fix(chat): remove unreachable Telegram-router restart wiring
2026-09-19 02:25:52 +10:00
sam 3a70a3a31a Merge pull request #875 from samcharles93/feat/archie-reactions-stream
feat(nats): provision the ARCHIE_REACTIONS fan-out stream
2026-09-19 02:22:29 +10:00
sam 6f0731c7da fix(chat): remove unreachable Telegram-router restart wiring
Review fix for PR #873. The previous attempt wired Router.Restart on the
Telegram router, but that path is unreachable: telegram.go:260 registers an
exact-match /restart handler (restartHandler, restart.go:24) which consumes the
update before the chat router is reached.

The wiring was therefore a false affordance: it made the Telegram router report
RestartAvailable: true for a callback no production path invokes, and the test
only passed because it assembled a LocalChatAdapter by hand rather than driving
the composition production uses.

Remove it. Telegram's own restartHandler remains the sole path to
RequestRestart. The removals of boot.restartTelegram and
telegramSetup.RegisterRestart stand: those were written and never read, which is
the real defect (channels-telegram-3).

Tests are retargeted and table-driven: buildTelegramRouter leaves Restart nil
with and without an adapter (regression guard against re-adding the unreachable
wiring), the router's /restart route reports not-configured, and
RequestRestart's queue-then-reject contract is pinned at the adapter level.

This deliberately does NOT fix dashboard /restart: the web router runs in the
archie-gateway process while the Telegram adapter runs in archied, and no
gateway->daemon restart RPC exists. That needs its own design (see the bead
referenced in the PR body).
2026-09-19 02:17:59 +10:00
sam 7b7401d98b fix(nats): bound the ARCHIE_REACTIONS stream by age
Review fix for PR #875. ARCHIE_REACTIONS is provisioned under LimitsPolicy,
which retains every message until a limit is reached. With no limit configured,
acknowledged reactions accumulated without bound and continuous traffic could
grow the stream until it exhausted disk.

nats.Config gains MaxAge *time.Duration, using the same pointer form as
Retention so "unset" (nil) is distinguishable from an explicit zero. It is
applied to the stream config only when set, so every existing caller keeps its
current behaviour.

ARCHIE_REACTIONS sets MaxAge = 24h. A reaction is a producer-only wake event
whose authoritative state is re-read at pass time, so a dropped reaction delays
work until the next poll (default 60s) rather than losing it; 24h tolerates a
consumer down for a full maintenance window and matches the package's existing
DefaultInactiveTTL.

ARCHIE_TASKS is deliberately left unbounded. It uses WorkQueuePolicy, which
discards a message on acknowledgement, so its size is bounded by outstanding
work; a limit there could discard a task that has not been delivered yet, which
is a worse failure than the disk risk. The table asserts that non-change.
2026-09-19 02:16:38 +10:00
sam 85b2724b2e test(archieui): make the shutdown regression test table-driven and bounded
Review fixes for PR #876. The production fix is unchanged; this reworks only
the test.

- Bound the wait for the handler-start signal and select on the request
  goroutine's error and on an early serve() exit, so a request that never
  reaches the handler fails promptly instead of hanging until the package-wide
  timeout.
- Restructure into a table, per AGENTS.md:256 / CLAUDE.md:256 ("table-driven
  tests against target behaviour"): a request that drains within
  ShutdownTimeout returns nil, and a request still in flight when it elapses is
  force-closed and still returns nil.

The overrun row was verified to fail against the previous
"return server.Shutdown(shutdownCtx)" and to pass against the approved
force-close. Both rows hold the request in flight deterministically, so no case
depends on machine load.
2026-09-19 02:12:27 +10:00
sam 5f97d8c976 fix(archieui): exit cleanly when graceful shutdown overruns its deadline
The gate failed intermittently with "task: Failed to run task \"test\": exit
status 1" and a tailed log that hid the failing package. The package was
cmd/archie-ui, which sorts before internal/logging and so never appeared in the
visible tail.

Cause: serve() returned the error from server.Shutdown, which is
context.DeadlineExceeded whenever a request is still draining when
ShutdownTimeout elapses. That converted a normal load-induced overrun into a
non-zero process exit, and TestUIProcessServesTheDashboardAgainstLiveDependencies
asserts a clean exit after SIGTERM.

Fix at the producer, in serve(): on overrun, force-close the remaining
connections and report a clean shutdown. No timeout is lengthened, no retry,
skip or -p 1 is added, and the smoke test's assertion is untouched.

Reproduced under parallel build load: -count=30 on the suspect test produced the
exact shutdown-deadline signature once; after the fix, 30/30 pass with zero
occurrences. The new test drives the deadline path deterministically by holding
a request in flight rather than relying on machine load.
2026-09-19 01:51:11 +10:00
sam 9e8427bc8e feat(nats): provision the ARCHIE_REACTIONS fan-out stream
archie.reaction.* had no matching stream. bootstrap.go:501-502 bound only
ARCHIE_TASKS to archie.task.> under WorkQueuePolicy, so any publish to a
reaction subject made js.PublishMsg fail and the publisher saw a publish error
-- the reason a previously landed reaction producer would have answered HTTP 500
on every webhook delivery.

ARCHIE_TASKS must stay WorkQueuePolicy: under it a second consumer on an
overlapping filter subject silently receives nothing, which is the trap recorded
in CLAUDE.md. Reactions need the opposite property, so they get a second stream:

- workintake.SubjectReactionWildcard ("archie.reaction.>")
- nats.DefaultReactionStreamName ("ARCHIE_REACTIONS")
- nats.FanOutRetention(), a pointer helper so composition can select
  LimitsPolicy without importing the NATS SDK, since nil Retention already means
  WorkQueuePolicy and the zero value (LimitsPolicy) is a valid explicit choice
- a second nats.Connect in connectNATS carrying that stream, closed on failure
  and registered for shutdown cleanup

Red-then-green: TestConnectNATSCreatesReactionStream failed with "publish to
archie.reaction.* = nats: no response from stream, want it to land in
ARCHIE_REACTIONS" before the stream existed.

Step 1 of docs/prds/pr-review-remediation.md (bead archie-core-8li9.1).
2026-09-19 01:51:07 +10:00
sam 9fd7f619bc fix(chat): wire the Telegram adapter restart callback into the chat router
/restart could never work in production. boot.setupGateways captured the
Telegram adapter's restart callback into boot.restartTelegram, a field no code
ever read, so gateway.Router.Restart was nil in every composition:
handleRestartAdapter always answered "not configured" and LocalChatAdapter's
snapshot always reported RestartAvailable: false.

Fix at the producer: buildTelegramRouter assigns Router.Restart directly from
the adapter it already holds. The dead field and the telegramSetup.RegisterRestart
indirection are deleted rather than kept alive.

Red-then-green: TestBuildTelegramRouterWiresRestartCallback failed with
"router.Restart = nil, want the telegram adapter restart callback wired into the
router" before the change. TestBuildTelegramRouterRestartUnconfigured pins the
unchanged no-adapter behaviour, including the not-configured reply.
2026-09-19 01:50:41 +10:00
sam cc390256a4 fix(config): enforce the service context at registration
ServiceSpec.Context documented which addresses a service must end up with, but
nothing read it, so a typo'd context was accepted and meant nothing. Review
caught it as scaffolding that looked load-bearing.

RegisterService now validates it: an unknown context, or a hosted (server/both)
service registered with no default listen address, panics. Registrations come
from this repository's own init(), never from operator input, so an invalid one
is a programming error with no caller positioned to handle an error return, and
failing at process start beats a service binding where nobody dials.

Services.RequireTarget replaces the same check and message hand-written in
composeStateStoreClient and composeChatContract. An empty registered default
target is what declares a target operator-supplied, so the requirement belongs
to the registry rather than to each consumer. It stays a lazy check at
composition time rather than becoming load-time validation: a process that
never dials a service must still load a configuration that omits it.

The review's stated failure mode does not occur. A hosted service with an empty
listen does not reach net.Listen as an ephemeral port; resolveServiceListen
returns "no listen address" and archied refuses to start. The defect was the
inert field, not a silent bind.

Refs: archie-core-u23g
2026-09-19 01:50:25 +10:00
sam 6bdbd5e31e feat(config): key services by name through a registry
Services was a struct with one field per service, so each service's name was
stated three times: the field, the loader's defaulting branch, and every
consumer's field access. Adding one meant editing all three, and nothing
stopped them disagreeing. The drift was already there: stateStoreResolvedToken
and gatewayResolvedToken were identical apart from the env var each hard-coded.

Services becomes map[string]ServiceConnection, and RegisterService declares a
service once with the five fields it has. Everything that differs between
services is now an argument rather than a branch elsewhere: an empty registered
target means the operator must supply one, which is how services.state.target
stays required while the gateway's is defaulted, and the registered token env
var replaces the two copied resolvers with Services.ResolvedToken.

Defaulting iterates the registry, so a new service needs no loader edit.
Consumers read Services.Get(name).

Keeping a typo'd section reported took explicit work. [services.gatway] used to
land in toml.MetaData.Undecoded() because the struct had no such field; a map
decodes any section, so it would have loaded silently and left the daemon
dialing the default. unregisteredServiceKeys asks the registry which names are
real, and the test for it fails with an empty UnknownKeys when that check is
removed.

The TOML shape is unchanged, so no deployment profile, config.example.toml or
install.sh edit is needed.

Refs: archie-core-u23g
2026-09-18 23:32:36 +10:00
sam e0beb68978 feat(config): let the State Store and Gateway listen addresses be configured
Both gRPC listeners took their bind address from a -listen flag default alone,
with no config key. The port a service binds and the port its clients dial were
therefore two separate facts -- the flag default, and [services.<name>].target --
with nothing keeping them in step.

On a host that already owned the default the effect read as a code fault rather
than a port collision: the store died with "bind: address already in use" while
every client reported "state store: connection error ... error reading server
preface: EOF". Cockpit binds 9090 by default, so this is not hypothetical.
Verifying the dashboard on such a host needed a hand-written overlay plus a
matching -listen, two things to keep in step by hand.

- [services.state].listen and [services.gateway].listen now name the bind
  address. Precedence is -listen, then config, then a default applied by the
  loader. An empty result is a startup ERROR, never a fallback: net.Listen reads
  "" as "any free port", so defaulting there would bind the service somewhere no
  client looks and still report success.
- Both flag defaults become "" so configuration can win, mirroring how archie-ui
  already prefers [web].listen over its flag.
- Gateway listen defaults to the same constant as the gateway target, so a
  default deployment's client and server cannot drift onto different ports.
  [services.state].target still has no default: an empty target stays a startup
  error in the daemon and gateway, which no longer serve a local store.
- ServiceConnection gains the field. Its doc now says clients read only Target
  and TargetToken, and Listen belongs to the process that owns the service --
  the previous wording claimed the struct starts no listeners while carrying one.
- deployments/dev.toml states each port once, beside the target it must match, so
  `task dev` is retargetable on a host that owns 9090 by editing one overlay.

Tests. Unit: config defaults, explicit values, and precedence. Process: the
existing real-process smoke test now takes its listen from configuration instead
of -listen, so the end-to-end wiring is proven by the test that already starts
the binary, with no extra process. Both were verified load-bearing by mutation --
inverting the precedence fails the unit test, and making the app layer ignore
config fails the process test with "state: no listen address".

Verified: task check exit 0. Also documented in config.example.toml, the
configuration architecture table, and the `task dev` summary.
2026-09-18 19:15:02 +10:00
sam 2ef962912a feat(tasks): attribute events, changes and config to the attempt that produced them
The events table had no attempt column, so two attempts of the same task were
indistinguishable: a retried task's stage events interleaved into one timeline and
nothing could say which run produced a given result.

- events gains an `attempt` column through the existing presence-gated
  migrateTasks loop rather than a version bump, so user_version stays 3 and an
  existing archie.db is migrated in place. A legacy-database test builds the
  pre-change schema and proves both that existing rows survive unattributed
  (attempt = 0) and that a new value round-trips.
- Event.attempt = 11 and ReadTaskLogRequest.stage = 9 are field additions to
  existing messages: no new RPC, no new table, and no change to any canonical error
  message that crosses the state-store wire.
- Every task event the workflow engine emits now carries the attempt. Producers
  that have no attempt yet (task_queued, binding_dispatch_failure) stay
  unattributed rather than guessing one.
- Changed files are captured as a durable changes_captured event at commit and push
  time, reusing the existing three-dot diff primitive: paths from FilePatches(),
  counts from Stats(), base_sha recorded as the merge base actually diffed rather
  than the base tip, a 200-file cap with totals kept complete. Consumption goes
  through an unexported optional interface so workflow.Trees -- the yaegi capability
  surface for interpreted stage plugins -- is not widened. A capture is also taken
  once the pull request exists, so the capture can link it.
- The per-attempt configuration is captured as a durable config_captured event
  carrying the non-secret config.TaskConfig, instead of a new table plus two new
  state-store RPCs: events are already this system's per-attempt provenance stream,
  so a table would have given one feature two provenance mechanisms. Redaction is
  proven against canary secrets. Measured payload: 539 bytes typical, 2333 bytes at
  20 models.
- Captures carry a schema marker and are refused unless their key set matches the
  declared payload shape, so drift surfaces as "recorded but could not be read"
  rather than as a capture of zeroes.
- Three read endpoints serve the attempt rail, the changed-file captures and a raw
  debug view, mapping to UI-owned response types rather than dumping store types.
- The task-log read gains a stage filter, bound to the stage logger for that
  stage's own lines; agent and tool output records no stage.

A task-scoped grant may already insert events on its own task, so this provenance
is exactly as trustworthy as the rest of that task's event stream. That limit is
documented rather than papered over.
2026-09-18 18:32:59 +10:00
sam 2c805ef15d feat(tools): wire the workspace codesearch index into grep
internal/tools/builtin.RegisterBuiltins has always accepted a GrepIndex and
nothing ever passed one, so every grep ran the direct walk no matter what
[indexing] was set to. internal/indexing was fully implemented and tested
with zero production callers, and internal/config decoded IndexingConfig into
nothing. Two halves of one defect.

The composition root now builds an indexing.Manager and hands it to the
workspace tool provider, which forwards it to RegisterBuiltins. deadcode
confirms the result rather than the commit message asserting it: unreachable
functions drop from 88 to 73, and all 15 internal/indexing entries are gone.

Failure degrades rather than propagates. A manager that cannot build logs a
warning and grep keeps its direct walk, which is exactly what it did before
this was wired. An accelerator must never stop the daemon starting.

The context is detached with WithoutCancel rather than replaced with
Background: the index build's lifetime is the process, while the gateway
path's boot context is cancelled once startup returns, so inheriting
cancellation would kill the build mid-flight. Background would have satisfied
the linter too, but would silently drop the caller's context values, so
registerTools now takes the context its callers already had.

internal/agentexec's RegisterBuiltins call is deliberately left alone. It
registers builtins only to harvest their names for a plugin tool-name
collision check and never executes them, so there is no grep there to
accelerate.
2026-09-18 13:40:45 +10:00
sam a484693f11 refactor(logging): pass the caller's context to TaskRegistry.Write
Write built its own context.Background() for the handler call, which meant
any new caller with a context to offer could not hand it over. Taking the
caller's context is what the next commit's park-reason writer needs, and it
is what contextcheck requires of a daemon-side writer.

No behaviour change: the context is used for the handler's own attribute
extraction and never consulted for cancellation -- a park is recorded at
exactly the moment a run's context has been cancelled, so a cancelled
context must not suppress the entry.
2026-09-18 01:49:42 +10:00
sam 353c9f7e17 fix(agentworker): hand the worktree back on every sandbox exit
A container run only restored ownership of the bind-mounted worktree on the
CommitAll/Push success path (archie-core#520). The agent runs as root inside
the container, so every other way a run can end -- a gate failure that parks
the task, a stage error, an early return -- left the worktree owned by UID 0
on the host. The daemon is the process that later cleans and resets that
directory as its own non-root user, and it cannot even unlink inside a
root-owned directory, so the next attempt failed before it could start no
matter what the worktree refresh did.

runTask now defers the same reconcileOwnership the commit and push paths
already call, so it runs at every exit. The retry recoveries in worktree
package cannot fix this from the daemon side: a non-root process cannot
chown, and removing the directory needs write access to it.

Observed through a supplementary group -- a non-root process may move a file
it owns into a group it is a member of -- so the test proves the chown ran
without needing root.

Residual, filed as archie-core-zcg8: a container that is SIGKILLed never
runs this defer.
2026-09-18 01:49:35 +10:00
sam e338473414 Merge pull request #846 from samcharles93/feat/835-disk-readiness-probe-watches-only-work-dir-so-a-full-root-fi
Disk readiness probe watches only work_dir, so a full root filesystem goes unreported (archie)
2026-09-17 22:39:15 +10:00
sam 4ef56edd77 Merge pull request #848 from samcharles93/feat/state-store-task-log-download
feat(staterpc): read task logs through the State Store and make them downloadable
2026-09-17 22:38:16 +10:00
sam 21d8250cde feat(webui): read a task's log through the State Store and let it be downloaded
The dashboard's attempt-log panel answered every read with "task logging is
optional and was not enabled for this run". That message was false: logging is
wired unconditionally in the daemon, and the log files exist. The dashboard
process simply had no reader -- composition left TaskLogs absent because no
contract carried it -- so the handler saw a nil registry and reported the
disabled shape, which the page rendered as a claim about configuration.

internal/logging now answers whether an attempt has a log at all: Tail sets
Result.Found, so "this attempt has no log" is a value rather than an empty
list indistinguishable from a read that never happened. Its registry gains
TaskLog (a decoded page) and TaskLogContent (the file verbatim, streamed,
because a log is unbounded input), and an absent reader reports
ErrTaskLogsUnavailable rather than an empty page -- a process that cannot read
logs must not implicate the attempt.

Those two answers travel the State Store contract as storecontract.TaskLogStore
(ReadTaskLog, StreamTaskLogContent) and are rehydrated on the client by
(Unavailable, canonical message), so archie-ui composes the reader from its own
store client instead of opening another process's files. The daemon's registry
is wired as the server-side reader in the State Store process only.

On top of that read, GET /api/tasks/{id}/logs/download serves the attempt's log
as a Content-Disposition attachment named task-<id>-attempt-<n>.log, and the
panel offers it. The JSON route keeps its shape and gains found, so the page
states the attempt has no log instead of blaming a setting.

Task log files stay owned by their directory's owner: archiving still removes a
task's log directory, and that removal is the daemon's, not the dashboard's.
2026-09-17 12:12:43 +00:00
samcharles93 8378c92739 Disk readiness probe watches only work_dir, so a full root filesystem goes unreported (archie)
Implements #835
2026-09-17 11:34:38 +00:00
sam b215636f99 Merge pull request #844 from samcharles93/fix/archied-watchdog-test-flake
fix(archied): stop the shutdown-watchdog test racing the scheduler
2026-09-17 21:24:29 +10:00
sam bd7fb053a7 fix(archied): stop the shutdown-watchdog test racing the scheduler
TestShutdownWatchdogHangingPastLeashForcesExit gave the watchdog two leash
lengths (60ms) to force an exit, but the 30ms leash only starts the timer --
the firing goroutine must then be scheduled before the injected exit is
observable. Measured margin is ~29.7ms, so the test is a wall-clock race
against the Go scheduler and fails intermittently on a loaded machine, where
it parks real work behind a gate failure unrelated to the change under test.

The assertion is now the contract that actually matters and that scheduling
jitter cannot falsify: the exit must not happen before the leash, and it must
happen. Waiting for it uses a generous ceiling, and the new elapsed >= leash
check catches an early fire strictly -- something the old test could not do,
because it only bounded lateness.

The inverse test asserts the ABSENCE of a forced exit, which load cannot
falsify (a slow runner only delays the exit being checked for), so it keeps a
tight bound and the suite stays fast.

Verified by mutation: removing the forced exit fails the first test, and
firing before the leash now fails it on the new elapsed check.
2026-09-17 11:17:19 +00:00
samcharles93 f93792d1c0 fix: baseline gate repair (go test ./... -count=1) 2026-09-17 11:03:29 +00:00
sam d565fb8ade refactor(forge): delete the unwired RepliesAfter surface
DELETE: IssueForge.RepliesAfter and its Reply type had no production caller --
the waiting_human status exists and is reachable, but automatic detection of a
human reply never landed, so the intended poller was never built. Neither
implementation paginated either (both read one page of 50), so the method
would have been wrong if wired as-is.

The manual path is untouched: waiting_human remains reachable and an operator
requeues through the existing task-action flow. Removing the interface method
also removes seven test fakes that existed only to satisfy it.

No behavioural test accompanies this: a pure deletion has no assertion to
make fail, and the compiler enforces completeness across every implementation.
2026-09-15 16:54:16 +10:00
sam 77d1e62f7d fix(memory): install the threat scanner the composition root never wired
WIRE, not delete: docs/architecture/configuration.md's behavioural-literals
inventory lists the "Memory safety scanner -- compiled prompt-injection and
sensitive-data patterns with fixed warn/block consequences" as current
behaviour, and internal/memory.Manager.HandleToolCall already refuses a write
whose content scans as ThreatBlock. The consumer existed; the producer never
called SetScanner, so Manager.ScanContent short-circuited to ThreatNone and
the gate was a permanent no-op.

setupMemory now installs memory.DefaultScanner. Consequences stay at the
documented defaults -- scanReject is false, so a prompt-injection match is
downgraded to a warn and logged rather than refusing the operator's write.

Red-then-green: TestSetupMemoryInstallsAThreatScanner asserts the composed
manager actually scans, not that a field was assigned; it failed with
"ScanContent = ThreatNone for a prompt-injection payload" before the change.
Mutation-verified by removing only the SetScanner call (API intact) and
watching the same assertion fail.
2026-09-15 12:49:13 +10:00
sam ac38c442d5 feat(channels): wire ConfigSchema/ValidateConfig into a startup validation loop
Every chat gateway (Telegram, email, webhook) implemented channels.Channel's
ConfigSchema/ValidateConfig contract but nothing ever called it, so a
misconfigured gateway would only fail once it tried to start rather than at
boot. setupEmailGateway/setupWebhookGateway now call ValidateConfig before
registering a start function and return false (fatal boot error, matching
setupTelegramGateway's existing convention) on failure; telegram_setup.go
does the same for the already-constructed *telegram.Gateway.

Fixed telegram.Gateway.ValidateConfig/ConfigSchema along the way: it only
recognised token_env, but chat.telegram.token (a secret reference) is an
equally valid, already-precedent-ordered credential source per
resolveTelegramToken. Wiring the old validator unmodified would have
rejected every Token-only deployment. Also corrected ConfigSchema's stale
allowed_chat_ids property name to the real allowed_user_ids toml key.
2026-09-15 11:22:23 +10:00
sam 6d5a1d7ff7 Merge branch 'crew/cluster3' into crew-integration 2026-09-15 10:33:15 +10:00
sam edb6373715 Merge branch 'crew/cluster2' into crew-integration 2026-09-15 10:33:15 +10:00
sam 2c80f3c59b Merge branch 'crew/cluster1' into crew-integration 2026-09-15 10:33:15 +10:00
sam 1fe6b9af48 feat(channels): wire chat.webhook route config through to the gateway
internal/channels/webhook's RouteConfig implements and tests HMAC
signature validation, payload template extraction, and reply delivery,
but setupWebhookGateway built []webhook.RouteConfig{{Path: "/webhook"}}
literally -- no config field could reach Secret/Template/DeliverTo, so
the gateway accepted any unsigned POST despite its package doc claiming
signature validation.

Adds config.ChatConfig.Webhook (WebhookRoute: path, secret as the
existing SecretRef pattern, template, deliver_to) and webhookRoutes(),
the pure translation into []webhook.RouteConfig that
setupWebhookGateway now calls. A secret ref that fails to resolve logs
and falls back to no validation rather than refusing to start, since
signature validation stays opt-in (an empty secret is documented,
valid config) unlike the mandatory forge webhook secret.
2026-09-15 10:22:54 +10:00
sam e13aeff59a refactor(archied): delete the daemon's unused webui.Server and UI adapters
b.web and its three adapters (skillCatalogAdapter, memoryUIAdapter,
curatorsUIAdapter) were constructed at boot but the daemon never calls
Handler() on the server and no route reads any of these fields -- the
dashboard is served by the separate archie-ui process (compose.go),
which deliberately leaves the same surfaces nil, and the daemon's
config snapshot goes through the standalone webui.BuildConfigView, not
a Server method. Removed the field, its four assignment sites, and the
three adapter files; test boot literals that carried a leftover
`web: &webui.Server{}` for no assertion-relevant reason had it
dropped, and the one test that specifically pinned b.web's store
surfaces staying nil is now moot with the field gone.
2026-09-15 10:19:00 +10:00
sam 0535b9ddd4 fix(channels): delete unreachable Telegram webhook delivery mode
telegram.New only ever constructed with empty webhookURL/webhookSecret
(telegram_setup.go) and config.TelegramConfig has no field to set them, so
the SetWebhook/StartWebhook branch, WebhookHandler, and webhook teardown in
Stop were dead code exercised only by tests. Long-polling is the only
reachable mode; delete the webhook path and its ConfigSchema properties so
the schema no longer advertises unsupported config.
2026-09-15 10:18:48 +10:00
sam 24e94828a3 refactor(staterpc): delete the superseded single-token auth interceptor
UnaryTokenInterceptor/TokenValidator/checkToken (auth.go) and
constantTimeTokenValidator (state_store.go) had no production caller:
server auth on the bridge-address listener is wired through
TaskGrants.UnaryInterceptor/StreamInterceptor, which replaced this
all-or-nothing token check with per-task scoped grants. Removed the
dead cluster and its now-pointless tests; TaskGrants' own auth
rejection paths are already covered in task_grants_test.go.
2026-09-15 10:18:25 +10:00
sam 72d9125bba fix(agentworker): wire configured tool-result policy into worker's LoopRunner
TaskConfig now carries the daemon's ToolPolicy (MaxResultChars/SpillDir)
via ForTask/ToConfig, and runTask applies it to the constructed
LoopRunner. Previously the worker always built the agent loop with a
zero-value ToolLimits, so tool results in every task's agent stage were
never capped or spilled despite the configured 50k default (agentexec-1,
tools-core-11).
2026-09-15 10:14:50 +10:00
sam 6ecde5cc24 feat(archied): wire the scheduling ticker engine at boot
schedulingConfig translated [scheduling] config into a domain
EngineConfig but nothing ever called it in production: the whole
scheduling/cron family (ticker engine, cronstore, crondelivery) was
built and tested but reachable from no binary, so an operator setting
scheduling.interval etc got a silently inert section.

setupScheduling opens a cronstore.Store as the engine's JobSource,
builds a crondelivery.Router with a workflow-kind runner backed by the
same non-forge gateway.TaskCreator chat's /spawn already uses, and
constructs the scheduling.Engine from schedulingConfig. startServices
starts it alongside every other subsystem; Stop is registered via
addCleanup on the same Background-timeout pattern as the curator
runtime. No chat task creator configured leaves the engine unstarted
rather than running with no reachable job kind -- the "chat" delivery
kind stays unmapped for the same reason (no unified cross-channel
outbound sender exists at this composition root yet).
2026-09-15 10:14:10 +10:00
sam ff01d7d335 test(archied): assert provider-disable warning by attribute, not message text
TestResolveProviderSecretsDegradesInsteadOfFailing matched a substring of the
warning's prose. Assert the structured provider attribute instead so a
rewording can't break the suite while a missing warning still fails it.
2026-09-15 05:25:12 +10:00
sam 2d3435e693 feat(archied): warn at boot when the config file has unrecognised keys
Surfaces configuration.Document.UnknownKeys as a warning naming the
offending keys. Stays a warning, not a boot failure, per
docs/architecture/configuration.md's startup policy: only an invalid
config is fatal, and a stray key is far more likely an operator typo in an
otherwise-working file.
2026-09-15 05:07:57 +10:00
sam 8de0df9ec3 test(archied): assert flush behaviour instead of grepping main.go
TestSubscribeAgentEventsFlushesBeforeReturning only checked that main.go's
source text contained the string "flush task event subscription", so it
passed whether or not subscribeAgentEvents actually flushed and failed if
the message were merely reworded. Replace it with a real embedded-NATS
test: subscribe, then publish on a separate connection with no delay, and
assert the event arrives -- repeated to make the SUB/PUB ordering race
observable. Mutation-verified: removing the flush call makes this fail
reliably (multiple iterations lose the race) while the text-grep version
would have kept passing.
2026-09-15 05:07:26 +10:00
sam 0e64573b04 refactor(install): install.sh delegates config generation, and tests stop reading hand-written configs
Two things fall out of one principle: a test that asserts against a
hand-written file is testing an artifact a human or an agent may legitimately
rewrite at any moment, so only generated output is a valid target.

install.sh generated config.toml two ways -- a full heredoc for the self-host
path and an awk splice of the [forge] block into config.example.toml -- with a
forge_block shell helper feeding both. That is the arrangement
docs/architecture/configuration.md decided against: it "must stop generating
config.toml; archied setup owns it", because the installer hand-rolling TOML
already drifted once (choosing Gitea produced a config naming a token the
installer never wrote). Both generators and forge_block are gone; install.sh
collects its answers as before and passes them to archied setup as parameters,
with secrets passed as REFERENCES to the env keys it just wrote rather than as
values, which would put them on a command line. The installer also now asks for
a cloud model, which archied setup requires and the old path never emitted.

deployment_contract_test.go tested the removed generators by parsing install.sh's
shell text, reconstructing the heredoc and asserting on its content. It broke the
instant install.sh stopped generating config -- "install.sh has no heredoc
declaring the daemon config's [nats] table" -- which is the failure mode in
miniature: the test was coupled to an artifact, not to behaviour. Deleted.

What it actually protected is kept, asserted on the GENERATED config instead:
TestRunSetup_GeneratedConfigSatisfiesDaemonRequirements runs archied setup and
loads what it wrote, requiring services.state.target and bot_user. A test against
generated output fails when generation breaks and passes when a human merely
tidies the template.

Found by an adversarial package sweep (temp tracker: config-1,
repo-root-deployment-1, repo-root-deployment-2).
2026-09-15 03:35:45 +10:00
sam a856e5ac80 fix(archied): disable a provider whose credential is missing instead of refusing to boot
configuration.md's startup policy -- "a missing credential disables a capability.
Only an invalid config stops the daemon" -- names this exact case: "Apply the same
reasoning to any future capability credential, LLM keys especially: start, report
the capability as unavailable, let the operator see it." The code did the
opposite. resolveProviderMap returned an error when a provider's api_key could not
be resolved, and again when it resolved empty, so configuredSecretRegistry failed
and the daemon would not start. A provider nobody referenced still blocked boot,
because the walk covers every entry in cfg.Providers rather than the ones [models]
uses.

The forge path was fixed to degrade for precisely this reason (resolveForge,
guarded by TestResolveForgeDegradesInsteadOfFailing); this is its provider
counterpart. A credential that cannot be resolved now disables that provider with
a warning naming it and why, and the daemon starts.

Disabling clears the reference rather than leaving it. webui's ProviderView
derives Configured from api_key or APIKeyEnv, so an unresolvable reference left in
place would advertise a provider nothing can use -- the "looks configured and is
not" shape this codebase keeps having to remove. Consumers already treat an empty
APIKeyEnv as no credential (agentexec/runtime skips such a provider, the daemon
exports nothing to the agent container), so clearing it is the whole disable.

The other half of the policy is kept: a reference naming only one of engine and
key is a config error, not a missing credential, and stays fatal.
applyForgeDefaults only ever builds a ref with both halves, so a half-named one can
only come from a hand-edited config.

This also defuses the trap 027dc99 introduced. archied setup can now write a
provider reference to any engine (-provider-secret-ref bws:OPENAI_API_KEY), and on
a machine without that engine the config used to load clean and then brick the
daemon. It now starts with that provider disabled and a warning.

The comment on templateDefaultActiveProvider described the old fatal behaviour and
is corrected rather than left to mislead the next reader.

All three behaviours are mutation-verified: returning an error again, accepting a
half-named reference, and disabling without clearing the reference each fail the
corresponding test. The red was an assertion failure, not a compile error -- the
logger parameter was threaded through first so the new test could be run against
unchanged behaviour.
2026-09-15 03:28:35 +10:00
sam 027dc9925b feat(configuration): let setup reference a secret without carrying its value
A parameterised install could not configure a cloud provider without a terminal:
the only way a secret value entered was the prompt, so an unattended run was
limited to the keyless self-hosted provider. The missing piece is the reference
form.

A reference is not a secret. An engine name and a key name are not sensitive --
which is exactly why a reference may be a parameter while a value may not -- so
setup gains -forge-secret-ref, -provider-secret-ref and -telegram-secret-ref,
each taking engine:key, parsed by one new setup.ParseSecretRef. When a reference
is given the step writes it to TOML and asks nothing, and setup neither reads nor
stores the value, because the value is not its to hold. That is the difference
from the prompt path, where the operator supplies a value and setup writes it
into the env engine and references it there.

Engine names are identifiers, not paths: the shipped engines are registered as
age, sops and vault, so a reference like vault/openbao:KEY is refused rather than
written into a config that could never resolve it.

Two refusals guard the silent no-op. A reference naming a key for the keyless
self-hosted provider, or for a forge that is disabled, would sit in the config
looking configured while nothing read it -- the failure shape this flow exists to
remove -- so both are refused rather than written.

telegram keeps writing token_env for the prompt path, because that form names an
env var and the prompt path is the env engine; a supplied reference may name any
engine, which token_env cannot express, so the reference path writes the
preferred token = {engine, key} form. Noted as the one asymmetry.

Also folded stepCloudProvider's key handling into cloudKeyEdit and cloudModel.
That was a lint fix for nested blocks, but the split is the honest shape: the two
branches obtain a key edit by different means and then resolve a model the same
way.

Every new test was mutation-verified by reintroducing the absence: ignoring the
forge reference, making ParseSecretRef validate nothing, and dropping the keyless
-provider guard each fail the corresponding test.
2026-09-15 03:23:08 +10:00