Replaces sourcegraph/jsonrpc2 with the go.lsp.dev stack (protocol, jsonrpc2, uri). The exit notification now ends Serve; closing the connection from inside the exit handler deadlocked.
Domain fields ParkClass/RemediationRounds landed in fe2a8e6d; this is
the wire and daemon half. Task gains park_class (32) and
remediation_rounds (33) and maps both directions; the classified park
write crosses the wire as ParkTask on a narrow TaskParker contract
(TaskLifecycle stays under the 8-method cap), with conformance
coverage local + grpc including the stale-guard sentinel and
needs_human normalization. Every daemon park site now states its
class: terminal for a repo that left config and a retired identity,
transient for the environmental parks (worktree, storage, grants,
taskrun), and identityMayAct returns the class beside its reason.
fe2a8e6d added Task.ParkClass and Task.RemediationRounds to the domain
without the wire half; the parity guard went red. Add the two proto
fields (32/33) and map both directions in values.go.
The F2+F3 schema wave from the auto-fix/resume investigation, sequenced
together as one migration. Every failure funneled to one parked status
with free-text ParkReason, and Task.RetryCount doubled as the
operator-retry budget and the review-remediation round budget -- so N
operator retries ate the remediation budget, a repo-level round cap
could re-park a task an operator had just legally retried, and no
surface could answer what a park needed from an operator.
Park class (F2): a producer-recorded vocabulary in taskstate --
needs_human (the default, so an unclassified park always reads as
operator-actionable), transient (environmental; a requeue can succeed),
terminal (retrying is pointless). Recorded at the park site, never
inferred from reason text; NormalizeParkClass makes unknown values
persist as needs_human. Transition-to-parked defaults the class, so
every existing unclassified park site reads correctly with no call-site
churn; the classified ParkTask write follows for parks that know
better. Requeue, RetryTask and BeginRemediation clear the class back to
the default with the reason.
Split budgets (F3): Task.RemediationRounds is the remediation's own
budget; RetryCount stays the operator's. The round-cap stage counts and
caps rounds; operator retry keeps counting RetryCount and can drive a
task to dead without touching rounds. remediation_rounds is a
presence-gated migrator column written through Update.
Known state: proto fields, the ParkTask RPC, the staterpc mapping and
the daemon-side park-site classification follow in the wire half of
this wave (the proto parity test is intentionally red until then).
Sequencing steps 6 and 7 of pr-review-remediation.md: the consumer
landed in 33e915c5, and these are the two sources that feed it.
Webhook (decision 2, low latency): the GitHub receiver decodes
pull_request_review and pull_request_review_comment events behind the
existing HMAC verification and publishes the same typed reaction the
poller produces, keyed source-independently so PublishUnique collapses
the two sources. Only submitted/created carry a verdict to remediate;
edited and dismissed are acknowledged. Reaction deliveries are
rate-limited per source (decision 6's 'add when the receiver gains a
second event family'), and the reaction path carries no dispatch
predicate -- eligibility is the consumer's owned-PR lookup.
Poll (both forges, the no-webhook backstop): one shared scan over all
pr_open tasks, each with the forge client its own identity owns
(reconcilePRs's shape), listing reviews and comments since per-task
persisted high-water marks and skipping the owning identity's bot user
(decision 5's own-comment exclusion, which the poll is positioned to
make identity-correct).
The cursors are two, not one: review IDs and comment IDs are independent
forge sequences, and a combined high-water mark would silently skip
reviews once a larger comment ID landed. watch_comment_id -- dead since
the comment-watch design it was created for never shipped -- carries the
comment cursor; review_cursor is a new presence-gated migrator column,
and SetReviewCursors is its guarded RPC (pr_open only, so cursors never
move under a run that owns the task). forge.ReviewComment gains ReviewID
so a comment is collectable into its review's unit (decision 5) on both
forges, and forge.Review gains Body so poll-delivered reviews carry
their summary.
Note: the store/proto/facade half of this work landed early in aac31b69
under a release session's message -- a shared-checkout collision, kept
rather than rewritten; that commit carries the review_cursor migration
and the SetReviewCursors contract, this one carries the producers.
Beads sync to a Dolt remote, so anything written into one is published.
Two beads filed earlier today named an operator host, its internal
address and its filesystem paths -- the tracker was treated as a private
store, which it is not.
The rule is stated where the tracker is described, next to the existing
one about .beads/ never reaching a commit, because it is the same class:
the tracker is not a private store. Host-specific facts belong in the
operator's memos instance, which a bead may point at.
The remediate workflow was registered but had no trigger: the reaction
stream was provisioned, the envelope, resolver and owned-PR lookup all
landed, yet nothing consumed the stream (pr-review-remediation.md's
sequencing left the consumer for this step). The daemon now drains the
reaction stream each cycle, before the task drain, so a reaction can be
claimed in the same pass that carried it.
Per decision 3, the store's owned-PR lookup is the authorization
boundary: an unmatched, merged or closed PR drops with a counter, and a
public reaction can never start an agent against a foreign repository.
Per decision 5, a review is the unit of remediation -- its comment
events are appended to the unit while it is queued and dropped once the
run is claimed (the round cap bounds the exchange); a standalone comment
is its own unit of one; approved reviews never remediate; and archie's
own comments, matched on the owning identity's bot user, are excluded so
its replies cannot retrigger it.
The guarded BeginRemediation/UpdateReviewPayload store methods are the
dedup: the pr_open -> queued transition carries the review unit in the
same write, so re-delivered reactions fail the status guard instead of
starting a second round. Both are new StateStoreService RPCs following
the Requeue shape, covered by the local/grpc conformance battery.
task check's fmt pass rewrites both, so the committed tree was not
fmt-canonical. Adopted verbatim, per the repository's rule that formatting
changes from the configured formatter are law.
Lane D. The collection existed in the event-capture schema with nothing
writing it; tool calls reached only the on-disk task log, readable as text
and never queryable.
The projection is a decorator on the State Store's task store, wired in
stateStoreDeps. InsertEvent persists the event first and only then, for a
tool_call event, writes one row. A projection failure is counted and logged
but never returned: event persistence is load-bearing, the transcript is
observability, and failing a task write because its shadow failed trades
the fact for the record of it. With no event-capture store the task store
passes through unwrapped.
The event adapts to the existing collection rather than the collection to
the event; args and duration_ms stay reserved. InsertToolCall stays silent
on the notify hook, matching the dispatch tables rather than the
operator-editable ones -- tool calls are high-frequency telemetry, not
dashboard activity.
The tool_calls collection existed in the event-capture schema but nothing
wrote it. Give it its write path:
- edastore gains InsertToolCall and the minimal TaskToolCalls read, writing
only the existing collection. No new index: the collection is an
append-only transcript and the SQLite predecessor had no unique index.
- internal/app/archied composes toolCallProjectingTaskStore, a decorator on
the task store that persists the event first, then -- only for tool_call
events -- writes one row. A projection failure is counted and logged but
never returned from InsertEvent: event persistence is load-bearing,
tool_calls is observability.
- stateStoreDeps wires the decorator around deps.Tasks when both stores are
open, and passes Tasks through untouched otherwise.
The event adapts to the existing schema, not the other way round: the tool
name comes from data, a succeeded call's detail is the row's result, a
failed call's detail its error; args and duration_ms stay reserved for a
producer that reports them.
Implements docs/prds/artifact-sender.md (Approved). archie-core now sends
generated artifacts to the offloaded.dev collaborative editor, whose
POST /api/artifacts route, bearer service token, PocketBase-backed
versions and idempotency semantics are the receiving contract.
- [artifacts] config section: base_url + token SecretRef with a
token_env fallback naming WORKSPACE_INGEST_TOKEN. An empty base_url
disables the sender entirely; a base_url with an unresolvable token
disables it with a warning, because a missing credential disables a
capability rather than stopping the daemon.
- internal/infrastructure/artifactsync: one-method Publish seam in the
Courier posture, full source per publish (versions of record are
snapshots), 503 retried three times with backoff, 401/400 surfaced
immediately with the field reason, deterministic request ids
(artifact:pr-review:<owner>/<repo>#<n>).
- ReviewReport gains a pure markdown renderer in internal/domain/workflow:
summary, one subsection per finding with the failure scenario and
disposition when present, verified properties as a checked-clean list.
- The daemon's PR reviewer publishes after a completed review when the
sender is configured, records the editor deep link on the result, and
degrades a failed publish to a log line -- publishing never fails the
producing work.
The cursor change renamed the field to 'cursor' as well as widening its
type, which buf breaking flags as FIELD_SAME_NAME and FIELD_SAME_JSON_NAME
on top of the type change. The name buys nothing: the field is already
documented as an opaque cursor, and the number and name are the wire
identity. Adopt the gate's verdict and change only the type, keeping the
break confined to the one rule already recorded in proto/buf.yaml.
Lane A. EventsSince no longer relies on events.id being monotonic: it
orders by (at, id), where at is written in a fixed-width layout so
lexicographic order is chronological order, and id is only a
deterministic tie-break. The cursor crosses the wire as an opaque string
and rides the SSE id: field, so EventSource's own Last-Event-ID resume
keeps working untouched. A malformed or legacy cursor degrades to
"from the beginning" rather than failing the feed.
events.id keeps its int64 type and row-identity meaning; no
push/subscribe RPC is added.
EventsSince no longer depends on events.id being monotonic. The feed now
pages a total order over (at, id): at is a fixed-width, lexicographically
sortable UTC timestamp (storecontract.EventCursorLayout) and id is only a
deterministic tie-break for rows sharing one timestamp. The cursor is an
opaque string on the wire (EventsSinceRequest.cursor), carried verbatim as
the SSE id: field and echoed back via Last-Event-ID, so EventSource resume
needs no client change.
A new idempotent migration normalises pre-existing RFC3339Nano at values to
the fixed-width layout before finishTaskMigration, and idx_events_at backs
the new query. Legacy integer and garbage cursors degrade to "from the
beginning", preserving the old failed-ParseInt-yields-0 behaviour.
Amends the "Dashboard live event delivery" decision's cursor-cannot-skip
justification, which previously rested on commit-order id assignment.
Lane C. A successful binding or mapping write on the State Store's
PocketBase tables now emits one activity event on the same task-DB path
capture arrival already uses, so a dashboard edit is visible on the
activity stream without a manual refresh.
The event is an invalidation signal, not a payload: it carries the record
id and the action, and consumers refetch through the authenticated API. A
nil notifier is a silent no-op, so stores composed without one (tests,
tooling) are unchanged.
Each successful write on the State Store's PocketBase bindings/mappings
tables emits one event (binding_changed / mapping_changed) through a
nil-safe Notify func injected at composition, so a dashboard edit is
visible without a manual refresh -- the same task-DB event path capture
arrival uses. Nil notifier keeps every existing constructor silent.
tools/cron ANDs the two day fields where Vixie cron ORs them, so
0 0 1 * 1 means Monday the 1st, not the 1st or any Monday. State the
deviation in the ScheduleCron comment and pin it with a test.
The event-capture tables (captures, mappings, bindings, binding_dispatches,
playbook_dispatches) now live in PocketBase-owned collections and their ids
are opaque record ids rather than autoincrement integers. The task lifecycle
stays on SQLite: tasks/transitions/events carry FIFO claim ordering, id-cursor
pagination and the task-grant map, none of which gain anything from being
hand-editable. binding_dispatches.task_id is the single seam and stays an
integer addressing that store.
BREAKING: CapturedEvent.Id, Mapping.Id, Binding.Id and Binding.MappingId are
strings on the wire, as are the ids in the capture/mapping/binding requests.
RecordDispatchRequest.task_id is deliberately unchanged.
What the collections buy, beyond storage: a payload nobody has modelled yet is
inspectable, a mapping is editable at runtime, and a binding's relation to its
mapping is enforced. The old schema declared no FOREIGN KEY on that link, so a
dangling mapping_id was previously possible and is now refused.
Behaviour preserved deliberately, each recovered from the SQLite version
rather than reimplemented from memory:
- draft -> pending_approval -> armed, with approval only from
pending_approval and an edit always returning a binding for re-approval
- one binding per source, guarded on insert, update and approve
- an empty secret on update keeps the stored one, so an edit form that does
not echo it back cannot silently disarm a binding
- ErrMappingNotFound / ErrBindingNotFound on a missing delete target
- sentinels are re-exported from storecontract, never redefined: staterpc
matches on exact canonical message to rehydrate them, so a second
definition would break errors.Is across the wire invisibly
Two implementation notes worth keeping:
- at-most-once dispatch is now a unique index rather than INSERT OR IGNORE.
Only the unique-violation maps to ErrAlreadyDispatched; any other save
error is returned, because treating them all as "already recorded" would
turn the ledger into a silent at-least-once dispatcher.
- captures store received_at themselves at nanosecond precision in a
fixed-width layout. PocketBase autodates are millisecond and record ids
are random, so same-millisecond captures would otherwise order
arbitrarily; the old store got insertion order free from autoincrement.
- ListUndispatchedCaptures is raw SQL: the anti-join is a subquery and
PocketBase's filter DSL is not SQL. A read fires no hooks either way.
Adds tool_calls, which had no persistence at all: tool calls only ever reached
the on-disk task log, readable as text but never queryable or watchable.
Nothing writes it yet.
The binding secret cipher moves with the bindings it protects.
The schedules feature no longer runs on it: jobs live in the state store
as the schedules control-plane document, read and written through the
control-plane client (scheduleResourceStore), and the legacy jobs.json
migration into that document has run its course. What remained of
internal/infrastructure/cronstore was 1,600 lines of file store, OS-level
cross-process locking and schema-versioned JSON envelope -- plus the
schedule vocabulary held hostage inside it.
The vocabulary moves to internal/domain/scheduling, which is where
migration-decisions already said a schedules vocabulary belongs: a domain
vocabulary package must not import infrastructure. JobSpec, Target,
Payload, the kind constants, Schedule and Duration move unchanged -- same
wire shape, same JSON tags -- with the two sentinel errors callers
matched outside the package (ErrJobNotFound, ErrScheduleUnsupported) and
ErrInvalidSpec, which the schedule arithmetic itself raises. The store's
own sentinels die with the store.
The legacy migration path (migrateLegacySchedules, readLegacyJobs) goes
with it: its only dependency was cronstore.Open, and its work is done.
crondelivery keeps its consumer-owned contracts; the compile-time
assertions that pinned *cronstore.Store move to the production
implementation, scheduleResourceStore, in internal/app/archied. The
package's fixtures swap the real file store for an in-memory one with
the same Get/MarkRun/Due semantics, so the run-recording acceptance test
still asserts against real schedule arithmetic.
A future storage backend (PocketBase) implements the same vocabulary.
First slice of the persistence migration, taking the event-capture tables
rather than the task lifecycle. These five tables hold no production rows and
their feature (archie-core-t2db) is unfinished, so they migrate at zero data
risk and the remaining work lands natively instead of being written against
hand-rolled SQL and ported later. They are also the tables that actually want
what PocketBase gives: inspecting a payload nobody has modelled yet, mapping
fields against it, and editing a binding at runtime are the epic's acceptance
criteria, not incidental conveniences.
The task lifecycle deliberately stays on SQLite. tasks/transitions/events
carry FIFO claim ordering, id-cursor pagination and the task-grant map, none
of which gain anything from being hand-editable. binding_dispatches.task_id
is the single seam and stays a plain integer addressing that store.
Idempotency moves from INSERT OR IGNORE on a composite primary key to unique
indexes, so the at-most-once ledgers are enforced by the schema rather than by
each caller remembering the pattern. A save that fails for any other reason is
returned rather than swallowed: treating every error as "already recorded"
would turn a broken ledger into a silent at-least-once dispatcher.
Adds tool_calls, which had no persistence at all. Tool calls were only written
to the on-disk task log (internal/logging), readable as text but never
queryable, filterable or watchable.
Both collection lookups treated any error as "not present". For the users
collection that returned nil on a real failure (caught by nilerr); for the
views it was worse but unflagged, because creating a fresh view collection on
the strength of a transient lookup error is a schema change made for the wrong
reason. Only sql.ErrNoRows now means absent.
PocketBase creates a "users" auth collection on first bootstrap whose
createRule is open, so an unauthenticated caller could POST a record and write
a row into the State Store's database file. This surface authenticates
superusers only and has no notion of end users, so the collection is deleted
rather than given a tighter rule: one that is absent cannot be loosened again
by a later PocketBase default.
Also re-applies the modernc pin. It was applied while developing the previous
commit but a concurrent go.mod resolution reverted it before that commit was
taken, so the tree has been running 1.59.0/1.77.1 against a PocketBase built
and tested on 1.57.0/1.74.4, printing the mismatch warning on every start.
Note for future dependency work: go get -u will undo this pin, and
modernc.org/libc does not follow semver, which is why PocketBase checks.
The claim in the previous commit message that the pin landed there is wrong.
Serve an optional PocketBase surface over archie.db from the process that
already owns it, behind archie-state-store -admin-addr. It co-tenants on
<db_path>-tasks.sqlite via PocketBase's DBConnect hook rather than opening a
database of its own, so the single-owner ownership claim covers it and
PRAGMA user_version stays archie's.
Every archie table is exposed as a view collection, which is read-only by
construction. Views alone do not hold that line: PocketBase also ships
POST /api/sql, which executes write statements as well as reads, and
/api/backups, which would put a second snapshot and restore semantics beside
store.Backup's VACUUM INTO. Both routes are refused rather than left behind a
superuser check, because a superuser writing task state is still a second
writer. The gRPC contract remains the only writer.
apis.NewRouter binds the JSON API but not the dashboard (apis.Serve mounts the
embedded UI itself), so /_/ is mounted here and the caller keeps the listener.
The dependency pin and the Taskfile/README changes for this landed with the
preceding commit, which swept up the staged files.
The index built google/codesearch over the chat workspace on every daemon
start, for the process lifetime, through a child-process helper and a
SQLite metadata DB -- and grep consulted it only for searches confined to
the workspace root, which the analysed sessions showed was a small
fraction of searches. The compute was continuous; the benefit was not.
Removes internal/indexing, the workspace-codesearch helper subcommand,
the [indexing] config section, and the GrepIndex accelerator seam from
the grep tool, which returns to its direct walk. google/codesearch
leaves go.mod with it.
A tool call rendered as a label line plus a ```text fence. Telegram draws a
fenced block as a full code widget (language header bar, copy button, panel),
and the block parser inserts a spacer paragraph before every block, so one
call cost four rendered elements. Three calls filled the screen to show three
word counts.
Every toolPreview path already summarises to a single line, so the fence was
never wrapping more than one. Drop it and put the preview on the header line.
Tool lines now join with a hard break rather than a bare newline: a soft
break would space-join them into one run-on paragraph, and a blank line would
put the spacer back. Three calls are now three consecutive blocks.
Two hardcoded values go with it. The status word is printed only when the
outcome is not success, since every line otherwise read "done" and
distinguished nothing. The icon is no longer a literal in the renderer: each
builtin declares its own on tools.Tool (not on Schema, which is the model's
contract), the provider carries it into ToolEntry, and the turn snapshots the
registry's icons alongside its tool set. A tool that declares none renders
without one rather than borrowing a generic wrench.
Carries a dependency upgrade (go.mod, go.sum, tools/go.mod, tools/go.sum).
CommonMark fences are three OR MORE backticks or tildes, and the closing
fence must be at least as long as the opening one. The parser hardcoded a
three-backtick prefix for both ends, so every longer fence was mangled:
````go language became "`go", matching no grammar
```` wrapping ``` the outer fence closed at the inner one and the code
leaked out as paragraph blocks
~~~go never recognised, and the "~~" emphasis pass then ate
two of the three tildes
The middle case is the damaging one: wrapping a reply in a longer fence is
how a model safely emits content that itself contains a fence, so getting it
wrong could wreck an entire reply rather than just its formatting.
Track the opening fence's character and run length, and close only on a run
of the same character that is at least as long and carries no info string.
CHANGELOG.archied.md and CHANGELOG.archie.md were two version streams that
drifted: a release that moved both components had two numbers, two pages and
two tags, and a reader had to visit two files to learn what shipped.
The merged history holds 65 versions from 64 archied and 34 archie sections --
33 that both components released, 31 only archied, one only archie -- so 98
published pages become 65. A version's date is the newest of its sections'
dates, and a component whose own date differs carries it on its label, because
the history has two dates and the merge must not invent one. `## [Unreleased]`
is not a release: it gets no page, and it stays at the top of the file.
A level-two heading is where the merge could silently drop a section, so the
merge is verified by content rather than by the generator: section counts per
source, the merged version set against the exact union, and every source `###`
title present both as a substring and, after stripping the component prefix, as
an identical heading. Counting `##` headings rather than releases overcounts by
one here, which is how the empty `[Unreleased]` block was caught.
- newsgen reads one file and writes `docs/news/<version>.md`, the index, the
canonical `releases.json`, and `docs/news/redirects.json`. The redirect table
is what keeps the retired `/news/<component>/<version>/` URLs alive; it is
generated beside the pages so it cannot drift from them.
- release.sh takes one `--version`, keeps commit and tag (the Woodpecker
pipeline that will own them does not exist yet), and exits 3 when the commit
set justifies no release, so a pipeline can finish green without tagging.
- The consumers of the two files follow: the deploy workflow's tag gate and
release-notes extraction, the zip contents, the update check's changelog
links, the packaged-path note, the docs skill, and newsgen's tests.
The per-component detail survives as labelled sections inside each release, so
a reader still learns whether archied or the agent changed; what goes away is
the separate numbers and the separate pages.
The live path recorded actor_id, actor_kind and principal_id as empty while the
actor was named inside the event's detail text. The daemon records the event and the
State Store holds it, and the contract between them did not carry the three fields,
so the columns were empty by the time anything could query them. An audit query
reading actor_id found nothing and "who approved this" meant parsing a
human-readable string -- the transient, unqueryable attribution this work exists to
replace.
The in-process test passed throughout because it used the local store and never
crossed that hop. The new test runs the same insert through the local adapter AND a
bufconn-backed gRPC client and asserts the columns on both, so the two paths cannot
disagree again.
Verified: task check exit 0 on this tree; the new test passes in both modes. The
grpc mode is reasoned to have failed before this change -- the field did not exist on
the contract, so it could not be set or read -- but that is inference from the
mechanism, not a run I performed against the old code.
Phase 1 of dynamic curator definitions (archie-core-trnd.1):
- ToolBuilder implementation resolves declared tool names from the
daemon's process-wide tool registry, returning exactly the declared
set and failing closed on an unknown name.
- curatorLLMRunner now carries ChatRequest.Tools and MaxSteps into
core.GenerateOptions and widens ChatResult with the executed tool
calls so a pass can attribute its actions.
- One generic DefinitionEngine interprets a Definition (Manifest +
name/enabled/instructions): its pass composes instructions with the
built declared tool set, runs the model loop, and records an action
per tool invocation.
- [[curators]] config definitions carry the Manifest shape verbatim plus
name/enabled/instructions; a definition missing interval is refused,
never defaulted.
- Idle curators are observable: the runtime logs a debug line when Check
reports not-due.
Two loose ends from the sign-in work, both mine.
A browser that cannot authenticate now goes to the provider when this instance has
a sign-in flow, instead of being shown the shared-token paste page -- a page that
cannot help it, because the token it would paste is not what this instance accepts.
A refusal that signing in again cannot fix, a suspended identity, is still answered
rather than redirected: sending someone back to a provider that already authenticated
them would hide the reason they were refused.
ProviderSession.Expired is removed. Nothing called it, and it must not: expiry is
the verifier's check, and a second opinion about whether a token is still good is
exactly the sort of near-duplicate that drifts from the authority it copies. The
browser path re-runs the flow when verification refuses, which is one rule rather
than two.
Verified: task check exit 0 on this tree.
The last gap in the identity boundary: enabling a provider used to lock browsers
out, because nothing could turn a sign-in into the credential the dashboard already
verifies. Now a browser completes the authorization code flow with PKCE and the
identity reaching the request context is the person who signed in.
The audience is requested AT THE AUTHORIZATION ENDPOINT, not only at the token
endpoint. Infra measured that a provider asked otherwise returns the audience empty,
and this dashboard then refuses a perfectly valid token -- a failure that reads as a
signing fault and is actually a missing query parameter. A test asserts the
parameter is present, so it cannot quietly go missing.
The flow is the libraries': discovery, the code exchange, PKCE generation and the
challenge transformation are go-oidc and x/oauth2, so archie writes no protocol. The
token returned by the exchange is verified through the SAME Verifier a presented
token goes through, and the identity is then resolved through the same rule -- one
Resolve now shared by Authenticate and the callback, so a browser sign-in cannot be
accepted on terms a presented token would be refused on.
Archie still issues no session. The callback stores the PROVIDER's token in an
HttpOnly cookie and every later request re-verifies it, so what archie holds cannot
outlive the provider's own decision. The callback checks the state it issued, and
expires the state and PKCE verifier cookies on use.
The sign-in routes mount on the bypass mux because they are how a caller becomes
authenticated; gating them behind the credential they obtain would make signing in
impossible. The client secret is never a flag: -oidc-client-secret-env names the
environment variable holding it, so it cannot reach a process listing or a log line.
No client id means no sign-in route and token verification is unaffected, which is
what an agent-only deployment wants.
Evidence, produced by driv·ng the flow rather than the seam:
internal/webui/api_tasks_actor_test.go's TestBrowserSignInRecordsTheSignedInPerson
walks sign-in -> callback -> cookie -> action and asserts the recorded actor is the
signed-in person with human_approved. TestCallbackRefusesAMismatchedState asserts a
replayed state is refused before any credential is verified.
internal/infrastructure/oidc/flow_test.go runs the real flow against a test
provider's own authorize and token endpoints, asserting the audience parameter, the
PKCE challenge, a successful exchange, and a refused wrong verifier.
Verified: task check exit 0 on this tree.
The updater is component-driven: it compares the daemon and agent versions and
acts only when they differ, so a host whose individual binaries disagree looks
healthy. Carina ran a hand-installed archie-messaging 1.35.0 against a 1.30.0
state store and crash-looped on an RPC the older store does not serve, while
the updater reported "no approved component changes; nothing to install".
The reason a comparison was impossible: three of the six host binaries carried
no version at all. archie-ui, archie-messaging and archie-playbooks did not link
internal/app/archied, so the release ldflags never reached them; the gateway and
state store linked the variable but exposed no flag. Only archied could say what
it was.
- internal/buildinfo carries Version, stamped by all three build sites: the
zip, install.sh, and the updater itself.
- Every host binary answers -version ("<name> <version>"). archied's existing
two-line output is unchanged, because something parses it.
- archie-update-install reads each installed binary before it can report
"nothing to install" and refuses, naming every binary that disagrees with the
daemon's version; after an install it verifies the same set against the
release it just installed, so a binary the release did not rebuild is skew
rather than success.
- TestDistZipShipsEveryHostCommand pins stamping across the three build sites
and requires each cmd/ to report, so a seventh command cannot arrive unstamped.
The transport could carry an actor and nothing filled it. It does now, so a
dashboard approval is recorded as the person whose credential verified, and an
agent's token is recorded as the agent's action -- for an agent, agent_approved;
for a person, human_approved; and for a caller whose identity archie could not
verify, task_approved, which claims nothing about who did it.
The actor travels the path the action takes: webui reads it from the request
context the middleware resolved, carries it in the gateway request, the daemon
passes it through the in-process adapter, and the NATS request body carries it to
the service that records it. Scope stays separate from actor at every hop, because
scope is which task a caller may touch and the actor is who touched it.
A chat channel's bound identity is an actor too: a bot acting on its own channel's
task is recorded as an agent's action rather than a person's, via ActorFromScope,
whose empty kind is what keeps it out of human_approved. The agent-facing
task_action tool passes that identity and never input["identity"].
The UI process's credential check is now filled by endpoint references rather than
by configuration: -oidc-issuer and -oidc-audience on archie-ui, two URLs the
process already understands the shape of, with the verifier built in Run because
cmd/ parses flags and nothing else. The UI package still never reads config, and a
configured provider that cannot be reached fails boot rather than falling back to
the gate that names nobody.
Evidence for this change, stated as measured: go build ./... and go vet ./... are
clean, and every package this touches passes -- webui, domain/taskactions,
infrastructure/taskactions, infrastructure/gatewayrpc, gateway, app/archieui,
app/archied, domain/messaging, channels/telegram, cmd/archie-ui. task check could
not be run to completion: it fails in cmd/archie-messaging on an untracked
internal/buildinfo package another lane has in flight, which is not this change and
was left untouched. The end-to-end flip is covered in
internal/webui/api_tasks_actor_test.go: a person's credential produces a record
naming that person with human_approved, an agent's produces agent_approved, an
unverified request produces task_approved with no actor, and a request carrying
forged actor fields cannot name one.
The noctx linter refuses httptest.NewRequest, which carries no context. The
requests these tests build are the ones an authenticated caller sends, so they
belong on a real context rather than a background one.
The verifier was a library nothing called. A request now carries a credential or
is refused: webui.Server.Authenticate resolves a presented bearer token to the
identity that may act, and requireIdentity attaches it to the request context.
With a provider configured that check replaces the shared token, so every request
resolves to a named identity or is refused; with none, the shared-token gate stays
for a loopback instance.
Nothing about the caller is read from the request. The subject comes from the
credential the provider signed and the identity is the record that subject is bound
to, which is what makes attribution possible at all: a caller cannot assert who it
is, it can only present what the provider gave it. A test sends an
X-Archie-Identity header and an actor field in the body alongside a valid token and
asserts the resolved identity is still the credential's.
Refusals follow HTTP's own distinction: absent or unverifiable credentials are 401
and name the scheme, while a credential that verified for an identity that may not
act is 403. Each is covered, because a refusal path with no test is a path nobody
has watched refuse.
The two state-store RPCs the dashboard needs to resolve a subject exist:
ResolveIdentitySubject and BindIdentitySubject, with the client implementing
identity.SubjectBinding (a narrower contract than Repository, so the administrative
facade keeps its shape) and the server answering Unavailable rather than pretending
when a process has no binding store.
chat.proto carries the actor for the next step: actor_id, actor_kind and
principal_id on ApplyOperatorTaskActionRequest, regenerated here. The transport can
now carry an attribution and nothing fills it yet, which is why the dashboard still
records task_approved rather than a human attribution.
Not wired in production yet, and worth stating plainly: dep.Authenticate is a seam
nothing sets, because the dashboard is served only by the UI process and that
process is documented to take endpoint references only -- there is deliberately no
config.Config in it. OIDC settings must therefore arrive as flags there, which is an
existing design boundary I did not want to cross unilaterally. config.Oidc exists
so the daemon-side wiring has somewhere to read from.
Verified: task check exit 0 after this commit; focused webui run records the
resolution and all four refusals.
install.sh built five of the six host commands, omitting archie-messaging, and
scripts/archie-update-install's guard only checked that every listed service had
a unit -- never that every host unit was listed. A v1.30.0 updater with the old
list therefore neither backed up nor restarted the messaging service, leaving a
hand-installed binary running under an otherwise updated host (archie-core-1c01,
archie-core-1faq).
- install.sh builds archie-messaging, and its summary names it.
- The updater refuses when a host unit this release ships is not in
GATEWAY_SERVICES, naming it, alongside the existing refusal for a listed
service with no unit.
- TestDistZipShipsEveryHostCommand -- the name deploy.yml:255 already promised --
now pins all four lists to cmd/: the zip's build loop, the zip's copy list,
install.sh's build loop, and the updater's GATEWAY_BINARIES/GATEWAY_SERVICES.
A fresh install still seeds only archied.service; that gap is tracked separately
as archie-core-enow.
Infra decoded real tokens from both clients, and two of the three facts changed
this code.
A machine token carries NO sub at all: it names its caller by client_id, and the
identity is that client. A person's token carries sub, the provider's stable
identifier for the person. The subject is now derived explicitly -- sub when
present, else client_id -- and a verified token asserting neither is refused
rather than accepted as an anonymous caller, because accepting it would let any
token the provider ever signed act as whoever the binding happened to hold. A
machine identity is therefore per client registration, one identity per client,
which is what makes per-agent identity a configuration decision rather than a code
change.
Scopes are read from either claim shape. The standard is a space-delimited scope
string and this provider also emits an scp array; reading only one shape yields no
scopes at all, which is worse than reading none deliberately.
The audience-at-authorisation-endpoint trap is pinned by a test: a token minted
without the requested audience arrives with aud empty, and the audience check
refuses it. That refusal is correct and the fix is the request the human flow
builds, not a weaker check here -- a test now fails if anyone weakens it.
The PRD records both as constraints: a person's client and an agent's client are
registered separately because a shared registration collapses the distinction the
attribution rests on, and a person's token requests the audience at the
authorisation endpoint as well as the token endpoint.
Verified: task check exit 0; internal/infrastructure/oidc passes, including
subject derivation for both token shapes and every refusal.
Two halves of one boundary: a credential resolves to an identity, and an action
records which identity acted and whose authority it used.
resource server. internal/infrastructure/oidc verifies a provider-issued bearer
token against the provider's published keys, using an OSS OIDC library: archie
compares no signature, validates no claim and rotates no key itself, holds no
client secret and issues no session, so a compromised archie cannot mint an
identity. Audience is required rather than optional -- a token minted for another
service must not authenticate here -- and discovery happens at construction, so an
unreachable provider is a startup failure rather than a request-time surprise.
identity.Authenticate is the single place a credential becomes an identity, and it
refuses in four distinguishable ways: no credential presented, a credential that
did not verify (wrong key, expired, wrong issuer, wrong audience, malformed), a
verified subject with no identity bound to it, and an identity that may not act.
The order is deliberate: verification happens before any field of the token is
used, so nothing an unverified token carries can select an identity. A suspended
identity is refused even though its credential verified, because suspension is
archie's decision and must not depend on the provider withdrawing a token.
The binding is the second bridge beside the legacy-name alias: identity_subjects
keys (issuer, subject) so the same subject string from another issuer is a
different caller, and rebinding moves a subject so two identities can never claim
one provider subject. The contract Authenticate needs is SubjectResolver, narrower
than Repository, so a consumer that only authenticates does not take the whole
identity CRUD surface -- and staterpc.Client, which asserts Repository, does not
need a new RPC to satisfy it.
attribution. taskactions now records two identities per action: the acting
identity and the authorising principal, both on the event row as new columns added
through the existing migration table. The rule that the bug turned on is enforced
by the event kind describing the actor: a person's approval is human_approved, an
agent's is agent_approved, and an action with no verified actor is task_approved,
a task-level event that credits nobody. An agent acting under a person's standing
approval records itself as the actor and that person as the principal, so
"approved by came from me" is answerable from the record rather than asserted.
Scope and actor are separated: scope decides which task a caller may touch and is
never promoted into an actor, because the dashboard's credential is a shared token
and possession of it is not evidence that a person was at the keyboard.
That last point is a deliberate fidelity trade, and the reason the transport is
not wired: the dashboard action path passes no actor yet, so dashboard approvals
are now recorded as task_approved rather than claimed as a human's. Recording
nothing false is the requirement; naming the human verifiably needs the dashboard
to authenticate a token and resolve it, which needs two new state-store RPCs
(ResolveIdentitySubject, BindIdentitySubject) -- the next piece, not a rewrite:
the verifier is a one-method interface, so introspection would be a second
implementation of it.
Verified: task check exit 0; focused runs recorded in the handoff. Opaque
authelia_at_ tokens were inspected rather than assumed and cannot be verified
against public keys; infra is configuring access_token_signed_response_alg RS256
on the archie client, with audience https://archie.catlow.cloud and issuer
https://auth.catlow.cloud.