Files
multica/server/pkg/db/queries/autopilot.sql
Bohan Jiang 4fac8d772f feat(attribution): Human Attribution Phase 1 (MUL-4302) (#5150)
* feat(attribution): Phase 1 foundation — provenance schema + resolver (MUL-4302)

Human Attribution, Phase 1 (地基) first increment. Every agent run must be
traceable to exactly one accountable human AND record at which waterfall level
that human was resolved, so a NULL originator can be told apart from a genuine
'no human in the chain'.

- migration 149: add originator_source (waterfall label) + delegation/retry/
  rerun/rule-version lineage + kind-tagged trigger evidence to agent_task_queue.
  No FK, no cascade, no CHECK on the source enum (MUL-4302 §7); nullable ADD
  COLUMNs = fast metadata-only change on the hot queue table.
- internal/attribution: the accountable-human vocabulary (Source, EvidenceKind,
  TriggerKind) + pure, unit-tested classification rules (ClassifyComment/
  ClassifyDirect). No DB, no authorization — provenance labeling only.
- service: attributionFor{IssueTask,TriggerComment} gather facts and delegate to
  the pure classifier; the legacy originator resolvers now delegate here so
  there is one source of truth. originator_user_id's VALUE is unchanged, so the
  Composio-overlay and canInvokeAgent A2A authorization boundaries are
  byte-for-byte preserved (MUL-4302 §1.3).
- enqueueIssueTask / enqueueMentionTask stamp originator_source + evidence;
  CreateRetryTask carries the parent attribution forward and records
  retry_of_task_id so retry and manual rerun stay separable (MUL-4302 §5).

Verified: go build ./..., go vet, gofmt clean; new attribution unit tests +
enqueue stamping integration test green; existing resolve_originator tests
unchanged.

Co-authored-by: multica-agent <github@multica.ai>

* feat(attribution): split accountable_user_id from originator, close enqueue bypasses (MUL-4302)

Phase 1, per Bohan's decision on the MUL-4302 thread: audit and authorization
answer different questions and get different columns.

- Migration 151 adds agent_task_queue.accountable_user_id (no FK, no cascade).
  Authorization keeps reading ONLY originator_user_id (canInvokeAgent A2A gate,
  Composio overlay); audit/UI/usage read accountable_user_id + source + evidence.
- Invariant (finalizeAttribution, single chokepoint + §11 tests): originator
  non-null ⟹ accountable equals it. The two diverge only when originator is null
  (autopilot / degraded fallback), which is the deferred rule_owner/owner_fallback
  increment; this lands the column + mirror-write so that split has a home.

- Close the NULL-source enqueue bypasses Elon flagged: chat, quick-create,
  deferred-fallback and run_only-autopilot now stamp originator_source + evidence
  (+ accountable where a human exists). Autopilot stays unattributed until the
  rule-version snapshot table lands, but is no longer a silent NULL-source row.
  Retry inherits accountable_user_id like the rest of the attribution lineage.

- Fix assign/promote attribution (§4): a member who assigns/promotes an existing
  issue is now the accountable human (and, by the invariant, originator) ahead of
  the issue creator. Threaded as an OPTIONAL actor override, so comment/rerun/
  autopilot paths keep today's resolution and create-with-assignee (creator ==
  actor) is unchanged. The squad leader gate already judged the same member.

Also merges origin/main: renumbers the attribution migration 149→150 (main took
149 for issue_origin_agent_create) and folds agent_create into ClassifyDirect's
origin inheritance.

go build/vet/gofmt clean; attribution unit tests + service stamp/actor tests +
handler suite pass on a fresh DB migrated through 151.

Co-authored-by: multica-agent <github@multica.ai>

* docs(attribution): fix accountable NULL semantics + close chat/quick-create evidence boundary (MUL-4302)

Addresses Elon's 2nd-round review on PR #5150 (pre-merge doc/evidence items):

- Migration 151 no longer overclaims NULL. accountable_user_id is NULL not only
  on pre-migration rows but on NEW rows whose audit source resolved no human yet
  (run_only autopilot writes originator_source='unattributed' with NULL
  accountable until rule_owner lands). Header + COMMENT ON COLUMN reworded so a
  schema reader does not misjudge the invariant.

- Chat now uses the UNIFORM evidence pair (kind=chat, ref=chat_session_id), like
  autopilot_run/issue_assignment, instead of relying only on the dedicated
  chat_session_id column — new EvidenceChat kind. Added a service test asserting
  chat stamps direct_human + chat evidence.

- Quick-create is documented as the ONE intentional no-antecedent-row path: no
  comment/issue/session/run exists at enqueue time (the run creates the issue), so
  trigger_evidence_kind/ref stay NULL while the human rides originator/accountable
  and source is direct_human — not a NULL-source bypass.

No authorization behavior change. attribution + service + handler suites pass on a
DB migrated through 151.

Co-authored-by: multica-agent <github@multica.ai>

* chore(attribution): renumber migrations 150/151 → 157/158 after merging main (MUL-4302)

main's #5162 ("unblock release migrations") renumbered the chat migrations and
took 150 (agent_task_coalesced_comments) and 151 (chat_read_cursor), colliding
with this branch's attribution migrations. Renumber them above main's new highest
(156) so TestMigrationNumericPrefixesStayUniqueAfterLegacySet passes:

- 150_agent_task_attribution      → 157_agent_task_attribution
- 151_agent_task_accountable_user → 158_agent_task_accountable_user

Fixed the internal "migration 150" references in 158's header to 157. Migrations
apply cleanly through 158 on a fresh DB; migration lint green.

Co-authored-by: multica-agent <github@multica.ai>

* feat(attribution): autopilot rule_owner — accountable = rule version publisher (MUL-4302)

Implements rule_owner (MUL-4302 §3.4), the first attribution source where the
accountable human diverges from the (NULL) authorization originator.

- Migration 159 adds the append-only autopilot_rule_version snapshot table (no FK,
  no cascade); migration 160 adds its CONCURRENTLY lookup index.
- Write-on-publish: CreateAutopilot appends v1 (publisher = creator); UpdateAutopilot
  appends a new version when a SUBSTANTIVE autopilot-row field changes (assignee /
  status / execution_mode) — cosmetic edits (title/description/template) write none.
  Both run inside the existing handler tx (atomic with the autopilot write).
- Dispatch resolution: both autopilot execution modes now resolve the active rule
  version and stamp originator_source='rule_owner', accountable_user_id=publisher,
  rule_version_id=<snapshot>, with originator_user_id left NULL (authorization
  unchanged). run_only stamps CreateAutopilotTask directly; create_issue resolves in
  attributionForIssueTask so both modes attribute identically. A missing version /
  non-member publisher degrades to unattributed — never fabricates a human.
- finalizeAttribution now enforces the invariant ONE-WAY: it mirrors originator onto
  accountable only when originator is valid, leaving an explicitly-set accountable
  (rule_owner / future owner_fallback) intact when originator is NULL. Added
  rule_version_id to CreateAgentTask so the create_issue path persists it too.

Also merges origin/main and renumbers this branch's attribution migrations
150/151 → 157/158 (main's #5162 took 150/151); rule_version table is 159/160.

Tests: attribution unit RuleOwner + one-way invariant table; service integration
tests proving an autopilot-origin issue stamps rule_owner + rule_version_id (and
degrades to unattributed with no version). Full service/attribution/handler/
migration suites pass on a DB migrated through 160; build/vet/gofmt clean.

Deferred (same PR): trigger-table republish (cron/webhook/event_filters) and
system-pause/archive versioning; owner_fallback + fail-closed; manual rerun.

Co-authored-by: multica-agent <github@multica.ai>

* fix(attribution): manual autopilot trigger → direct_human to the triggering member (MUL-4302)

Elon's blocking finding: a member manually triggering an autopilot was attributed
rule_owner (accountable = rule publisher, originator NULL) like a schedule/webhook
run, so member B triggering member A's autopilot landed accountable=A and carried
no originator authorization context for the run. Per MUL-4302 §4 a manual "run now"
is a direct human action and must attribute direct_human to the triggering member.

- Thread the triggering member from TriggerAutopilot into dispatch: new
  DispatchAutopilotManual carries actorUserID (resolved via resolveActor +
  memberActorUserID, so only a member actor is a human; an A2A agent actor falls
  back to rule_owner). DispatchAutopilot / DispatchAutopilotForPlan keep their
  public signatures (pass an invalid actor); only the internal dispatchAutopilot /
  dispatchCreateIssue / dispatchRunOnly gained the param, so the many existing
  callers are untouched.
- run_only: dispatchRunOnly stamps direct_human (originator == accountable ==
  actor, no rule_version) for a manual actor, else rule_owner. CreateAutopilotTask
  gains an originator_user_id param for the manual case.
- create_issue: dispatchCreateIssue enqueues a manual trigger via the actor-carrying
  *WithHandoff entry points; attributionForIssueTask's autopilot-origin rule_owner
  branch is now guarded on !actorUserID.Valid, so a valid actor falls through to the
  direct_human override. Both execution modes attribute identically.
- schedule / webhook keep rule_owner (no actor). Trigger-table + system-pause/archive
  versioning remain the pre-merge follow-ups.

Tests: the run_only row assertion Elon asked for (schedule → rule_owner row on
CreateAutopilotTask), plus manual direct_human on BOTH modes (run_only and
create_issue), including a manual actor distinct from the rule publisher. Full
service/attribution/handler/migration/scheduler/cmd suites pass on a DB migrated
through 160; build/vet/gofmt clean.

Co-authored-by: multica-agent <github@multica.ai>

* feat(attribution): owner_fallback + fail-closed policy, and manual-rerun direct_human (MUL-4302)

Two of the three remaining Phase 1 items (trigger-table / system-pause versioning
is deferred — see PR description).

owner_fallback + fail-closed (§1/§3.5) — the never-null accountable guarantee:
- attribution.OwnerFallback degrades an UNATTRIBUTED result to owner_fallback:
  accountable = agent owner, originator stays NULL (audit-only, authz untouched),
  Source.Precise()==false. finalizeAttribution's one-way invariant already allows
  accountable-set / originator-NULL divergence, so nothing else changes.
- Migration 161 adds workspace.attribution_fail_closed (default FALSE) + a lean
  GetWorkspaceAttributionFailClosed read. (Also added the column to ListWorkspaces'
  explicit column list so its row type stays db.Workspace.)
- applyAttributionFallback is applied at every enqueue boundary (issue, mention,
  chat, quick-create, deferred-fallback, autopilot run_only): unattributed →
  owner_fallback (agent owner) by default, or ErrAttributionFailClosed when the
  workspace is fail-closed, which the caller surfaces to refuse the enqueue (the
  run does not start). So no run is left without an accountable human, and a
  compliance workspace can block unattributable runs instead.

manual rerun (§5) — a rerun is a NEW direct_human trigger to the rerunning member:
- RerunIssue threads the acting member (resolved in the handler via resolveActor)
  down to enqueueRerunTask, and attributionForIssueTask is now actor-first so the
  actor wins over an INHERITED trigger comment (a rerun keeps the comment for the
  daemon's prompt context but must attribute to whoever clicked rerun, not the
  original comment's human).
- rerun_of_task_id lineage is recorded via a targeted SetAgentTaskRerunOf update on
  the rerun path only (keeping the shared CreateAgentTask insert untouched), so
  system retry (retry_of_task_id) and human rerun stay separable in reporting.

Tests: OwnerFallback unit test; owner_fallback + fail-closed-refusal + manual-rerun
(direct_human + rerun_of_task_id) service tests; the prior "degrades to unattributed"
test updated to owner_fallback. Full service/attribution/handler/migration/scheduler/
cmd suites pass on a DB migrated through 161; build/vet/gofmt clean.

Co-authored-by: multica-agent <github@multica.ai>

* fix(attribution): close fail-open holes + move rerun_of_task_id into creation snapshot (MUL-4302)

Addresses Elon's two must-fixes on PR #5150.

1. accountable-never-null / fail-closed had fail-open holes. applyAttributionFallback
   now, for an UNATTRIBUTED run, refuses the enqueue (ErrAttributionFailClosed) in
   THREE cases instead of silently degrading to a runnable NULL-accountable task:
   - workspace policy read fails (or no workspace) → fail closed; we cannot confirm
     fallback is permitted, so we don't run an unattributable task on a DB hiccup.
     (Only the rare unattributed path pays this; precise runs never read the policy.)
   - workspace is fail-closed → refuse (unchanged).
   - owner_fallback has no valid agent owner → refuse rather than enqueue a task with
     a NULL accountable_user_id.
   ErrAttributionFailClosed's doc now covers all three "cannot guarantee an
   accountable human" refusals. Added missing-owner / policy-read-failure /
   precise-passthrough tests.

2. manual rerun rerun_of_task_id was a post-notify UPDATE (race: the queued event /
   daemon claim could see rerun_of_task_id = NULL, and a failed update degraded the
   run to a plain direct_human). It now rides the CreateAgentTask insert — threaded
   through enqueueIssueTask / enqueueMentionTask as a creation param (like
   retry_of_task_id) so it is written in the same statement before the daemon is
   notified. Removed the SetAgentTaskRerunOf follow-up query.

Also merges origin/main (unrelated CLI fix #5167, no conflict). Full service /
attribution / handler / migration / scheduler / cmd suites pass on a DB migrated
through 161; build / vet / gofmt clean.

Co-authored-by: multica-agent <github@multica.ai>

* feat(attribution): rule_owner versioning on trigger edits + system-pause/archive (MUL-4302)

The final remaining Phase 1 item: substantive publishes beyond the autopilot row now
republish the rule version, so a run's rule_owner accountable follows whoever last
changed what the rule does.

- Extracted the config-summary + insert into service.RecordAutopilotRuleVersion so
  the handler and the (different-package) failure monitor share one writer; the
  handler's recordAutopilotRuleVersion is now a thin wrapper.
- Trigger edits: UpdateAutopilotTrigger and DeleteAutopilotTrigger republish the rule
  version with the acting member as publisher, ATOMICALLY (tx-wrapped mutation +
  version write, mirroring CreateAutopilot/UpdateAutopilot). CreateAutopilotTrigger
  republishes best-effort — the webhook path mints its token with a retry loop that
  cannot share one tx, and a create is usually initial setup already covered by v1;
  a failed write there is benign (active version stays the current publisher, the new
  trigger fires under it, no immediate daemon claim rides it).
- Archive (DeleteAutopilot) republishes (member, status=archived), tx-wrapped.
- System auto-pause (failure monitor) republishes with a 'system' publisher,
  best-effort — a background sweep to a non-dispatching state (a paused autopilot
  never dispatches; a later member resume supersedes).
- RotateWebhookToken / SetSigningSecret deliberately do NOT version: they rotate
  credentials, not the rule's behavior (not §3.4 substantive).

Semantics: a system-published (no-member) active version degrades dispatch to
unattributed → owner_fallback, never fabricating a human.

Tests: republish-reattributes (member A → member B supersedes → dispatch resolves to
B; system publisher → unattributed). Full service/attribution/handler/migration/
scheduler/cmd suites pass on a DB migrated through 161; build/vet/gofmt clean.
Also merges origin/main (unrelated frontend feature #5074).

Co-authored-by: multica-agent <github@multica.ai>

* fix(attribution): make trigger-create rule-version republish atomic (MUL-4302)

Addresses Elon's final Phase 1 blocking finding: CreateAutopilotTrigger recorded the
rule-version republish best-effort AFTER the trigger insert. If member B added a
schedule/webhook trigger to member A's autopilot and the version write failed, future
schedule/webhook dispatches would keep attributing to A — violating the rule_owner
invariant that the last member to substantively change the rule owns future runs
("no immediate daemon claim" doesn't save it, since the miss surfaces at the LATER
trigger firing).

Both create paths now write the version in the SAME tx as the trigger INSERT:
- schedule create: wrap CreateAutopilotTrigger + recordAutopilotRuleVersion in one tx.
- webhook create: each mint-with-retry attempt runs in its own tx (insert + version
  commit together; a token collision rolls that attempt back and retries with a fresh
  token; a version-write failure rolls the trigger back). Passes ap + the acting
  member id into the helper.
- removed the best-effort recordTriggerRuleVersionBestEffort helper (and the now-unused
  slog import).

Test: TestCreateTrigger_RepublishesRuleVersionAtomically drives both create paths
through the handler and asserts a rule version is published by the acting member.
Existing webhook/trigger/archive handler tests still pass. Also merges origin/main
(unrelated avatar feature #5074). Full service/attribution/handler/migration/
scheduler/cmd suites pass on a DB migrated through 161; build/vet/gofmt clean.

Co-authored-by: multica-agent <github@multica.ai>

* feat(attribution): Phase 2.1 — surface run attribution on the task API (MUL-4302 §9)

First Phase 2 (visibility) increment: the agent-task API now returns the resolved
accountable-human provenance so the UI can render an "on behalf of" badge.

- AgentTaskResponse gains an `attribution` object: source label (never blank —
  pre-migration NULL renders "unattributed") + `precise` flag (false for the degraded
  owner_fallback / backfill / unattributed sources), the initiator (accountable) and
  originator (authorization) user refs, the evidence {kind, ref_id} pointer, and the
  rule_version / delegated / retry / rerun lineage ids.
- The label + evidence + raw ids are built in the PURE taskToResponse (no DB), so
  every task response carries them. Names are hydrated separately, only on the
  user-facing surfaces (ListAgentTasks, ListWorkspaceAgentTaskSnapshot, RerunIssue,
  CancelTaskByUser) — daemon-claim paths stay lean.
- Hydration resolves initiator/originator from the GLOBAL user table (departed-member
  safe) via a new batch GetUsersByIDs query (no N+1); best-effort, so a lookup hiccup
  leaves the raw ids intact.

Tests: pure taskAttributionBase (direct_human / rule_owner NULL-originator /
owner_fallback degraded / pre-migration→unattributed) + DB hydration (fills known
ref, leaves unknown id un-filled, skips nil). Full handler/service/attribution/
migration/scheduler/cmd suites pass on a DB migrated through 161; build/vet/gofmt
clean. The field is additive — the frontend's parseWithFallback ignores unknown keys,
so nothing breaks until the UI increment consumes it.

Also merges origin/main (unrelated editor feature #5090).

Remaining Phase 2 (next increments, same PR): frontend zod schema + "on behalf of"
badge + evidence-chain jump; append-only correction events (write + display).

Co-authored-by: multica-agent <github@multica.ai>

* feat(attribution): Phase 2.2 — on-behalf-of badge in the execution log (MUL-4302 §9)

Surface the accountable human on every agent run row:
- AttributionBadge composes Badge + ActorAvatar, shows "on behalf of <member>"
  with the resolution source as a tooltip; degraded (non-precise) attribution
  gets a warning tone, and an unresolved initiator renders an explicit
  "no responsible member" chip.
- Wire the badge into both active and past rows of the execution log.
- Mirror the attribution shape into AgentTaskResponseSchema (defensive, .loose())
  so the cancel-task path carries it through zod; add parse tests.
- Export TaskAttribution/AttributionUser/TaskEvidence from @multica/core/types
  and add the attribution block to all four issues.json locales.

Co-authored-by: multica-agent <github@multica.ai>

* fix(attribution): hydrate initiator names on issue-facing task endpoints + bound the badge (MUL-4302 §9)

Address Elon's PR #5150 review:
- ListTasksByIssue (the execution-log data source), GetActiveTaskForIssue and the
  issue-scoped CancelTask now call hydrateTaskAttributions, so the "on behalf of
  <member>" badge shows the real member name on issue detail instead of falling
  back to "someone". Mirrors the existing ListAgentTasks / snapshot behavior.
- AttributionBadge: cap width (max-w-40, min-w-0) and truncate the name span so a
  long name / narrow right column can't squeeze out trigger/status/actions; keep
  the avatar shrink-0.
- Add a handler test asserting the issue task list returns a hydrated
  attribution.initiator.name.

Co-authored-by: multica-agent <github@multica.ai>

* fix(attribution): use semantic AvatarSize 'xs' for the badge avatar

main refactored ActorAvatar.size from a raw pixel number to the semantic
AvatarSize union (packages/ui/lib/avatar-size). Switch the on-behalf-of badge
avatar from size={14} to size="xs" (16px) after merging main.

Co-authored-by: multica-agent <github@multica.ai>

* fix(attribution): stage-cascade falls back to parent-issue provenance, not agent owner (MUL-4302)

When closing the last sub-issue in a Stage wakes the parent's assignee agent, the
run was enqueued via a system-authored child-done comment with no actor, which the
resolver classified as unattributed and then degraded to owner_fallback (the agent's
own owner). That is the wrong accountable human: the woken run should be accountable
to whoever caused the parent issue to exist.

attributionForIssueTask now detects a system-authored trigger comment and falls
through to the parent issue's own provenance — the same creator / agent_create-origin
/ autopilot-origin chain a direct enqueue resolves (so an agent-decomposed parent
attributes via delegation to the human who drove it; a member-created parent to that
member; an autopilot parent to the rule publisher). owner_fallback is now only the
last resort when the parent provenance itself has no human.

- Extract attributionFromComment so attributionForIssueTask can inspect author_type
  without a second GetComment; authorization resolution stays byte-identical.
- Add a DB-backed test asserting a system child-done comment resolves to the parent
  issue's origin human (delegation), not owner_fallback.

Co-authored-by: multica-agent <github@multica.ai>

* feat(attribution): autopilot runs attribute to the firing trigger's creator (MUL-4302)

Per Bohan: an autopilot schedule/webhook run should be accountable to the human
who created the SPECIFIC trigger that fired it, not the rule publisher. (Manual
triggers already attribute to the invoking member via direct_human — unchanged.)

- Migration 162: add autopilot_trigger.created_by_type/created_by_id (nullable, no
  FK/cascade). Capture the creating member at both trigger-create sites (schedule +
  webhook).
- New precise source trigger_owner: originator stays NULL (an autonomous fire
  carries no human authorization — same authz-safe divergence as rule_owner),
  accountable = the trigger's member creator.
- triggerOwnerAttribution resolves run.trigger_id → creator; wired into run_only
  dispatch and the create_issue path (bridging issue → active run → trigger_id).
  Legacy triggers with no recorded creator, and agent-created triggers, degrade to
  rule_owner then owner_fallback — nothing regresses.
- Frontend: trigger_owner source label in all four locales + badge switch case.
- Tests: attribution TriggerOwner unit + Precise/invariant; DB-backed resolver
  tests (member creator → trigger_owner; creatorless → rule_owner fallback).

Co-authored-by: multica-agent <github@multica.ai>

* chore(attribution): re-trigger CI (dropped synchronize event on 249090260)

Co-authored-by: multica-agent <github@multica.ai>

* fix(attribution): mirror accountable_user_id on comment-coalescing merge (MUL-4302)

The one-way invariant is 'originator_user_id IS NOT NULL ⟹ accountable_user_id =
originator_user_id', enforced at finalizeAttribution for enqueues and preserved by
the retry-clone (copies both columns). But MergeCommentIntoPendingTask (main #5192)
re-stamps originator_user_id to the newly-coalesced comment's human WITHOUT touching
accountable_user_id — so folding member B's comment into member A's queued task left
originator=B / accountable=A, violating the invariant. Re-stamp accountable to mirror
the new originator (same thing finalizeAttribution does). Add a DB-backed regression
test.

Co-authored-by: multica-agent <github@multica.ai>

* fix(attribution): Elon's 3 must-fixes + DB cross-column invariant CHECK (MUL-4302)

Bohan approved Elon's plan; this closes the three attribution boundaries he flagged
and locks the one-way invariant at the DB.

1. Delegation now inherits the parent's ACCOUNTABLE, not just its originator. An
   autopilot-rooted chain (parent originator NULL, accountable = trigger creator)
   @mentioning an agent / creating a sub-issue used to drop to unattributed →
   owner_fallback and fail-closed workspaces wrongly rejected the fan-out. Added
   ParentAccountable/OriginAccountable to CommentFacts/DirectFacts; ClassifyComment/
   ClassifyDirect copy accountable down (source=delegation, precise, originator NULL)
   so the chain root stays stable at any depth (§3.2).
2. The direct-chat send path (SendDirectChatMessage, MUL-4351) wrote only
   originator_user_id — no accountable/source/evidence, a NULL-source bypass. It now
   stamps the full direct_human attribution like EnqueueChatTask.
3. Comment-coalescing merge re-attribution is now ATOMIC: MergeCommentIntoPendingTask
   re-stamps the whole snapshot (person columns + source + delegation lineage + rule
   version + evidence) of the new comment, not just the two person fields, so a merged
   run never shows B accountable while pointing at A's stale source/evidence.
4. Migration 169: NOT VALID CHECK (originator_user_id IS NULL OR (accountable_user_id
   IS NOT NULL AND accountable_user_id = originator_user_id)). Enforces the invariant
   on every new write (the class of bug #5192 introduced); historical rows not blocked,
   VALIDATE after Phase 3 backfill. Updated test fixtures that seeded originator-only
   rows to also set accountable.

Merged latest main (renumbered attribution migrations 163–168 after main took 161/162;
merged the retry-clone chat_input_task_id + attribution columns). Verified: go build/
vet, attribution/service/handler/cmd-server tests on a migrated DB, frontend
typecheck/lint/tests.

Co-authored-by: multica-agent <github@multica.ai>

* fix(attribution): trigger responsibility transfers to effective publisher; fail-closed comment merge

Elon final-review must-fix 1 (MUL-4302): trigger_owner now resolves to the
member CURRENTLY responsible for the firing trigger's effective config, not the
fixed creator. Per-trigger published_by on autopilot_trigger, seeded to the
creator and re-stamped to the editor on a substantive edit — a trigger-scoped
edit bumps only that row (UpdateAutopilotTrigger), an autopilot-level edit bumps
all its triggers (UpdateAutopilot). Editing one trigger never reassigns another.
Adds real dispatchRunOnly transfer test + resolver-level isolation test.

Must-fix 3: AttributionForMergedComment reuses applyAttributionFallback and
returns ErrAttributionFailClosed; the merge caller refuses on fail-closed,
keeping the queued task's original precise snapshot instead of degrading it to
owner_fallback. Adds regression test.

Renumbered attribution migrations to 166-172 after merging main.

Co-authored-by: multica-agent <github@multica.ai>

* fix(attribution): make migration 172 invariant CHECK upgrade-safe (Option A)

Elon final-review must-fix 2 (MUL-4302), rollout chosen by Bohan. The NOT VALID
CHECK still checks a pre-existing row on any later UPDATE (even one not touching
attribution columns), so cross-deployment stale queued/running tasks (originator
set, accountable NULL from before migration 167) would fail on their next
claim/complete/cancel. Exempt exactly those legacy rows via 'originator_source
IS NULL' — that column was added in 166 with no backfill, so it is NULL only on
pre-migration rows and non-NULL on every attribution-aware write. New writes
stay fully enforced; the #5192 bypass class (source always set) is unaffected.
Phase 3 backfills legacy rows then drops+re-adds the strict form + VALIDATE.

Adds TestAttributionInvariantCheck_ExemptsLegacyRows (legacy row survives a
status UPDATE) and updates the reject-bypass test to the enforced regime.

Co-authored-by: multica-agent <github@multica.ai>

* fix(attribution): pin substantive/cosmetic edit boundary for trigger transfer

Elon re-review must-fix 1: an autopilot-level edit only transfers trigger_owner
responsibility when it changes what the automation instructs or who/whether it
runs. autopilotRuleSubstantiveChange now includes description (the run Prompt)
and issue_title_template; title and project_id stay cosmetic/routing.
UpdateAutopilotTrigger no longer transfers on every PATCH — it compares the
persisted before/after and transfers only on a real cron/timezone/enabled/
event_filters change, not a label-only or no-op PATCH. Adds real handler tests
(prompt->all, title->none, cron->one+isolation, label->none, no-op->none).

Must-fix 3: real merge-path regression (TestMergeCommentIntoPendingTask_
FailClosedKeepsOriginalSnapshot) drives mergeCommentIntoPendingTask and asserts
a fail-closed workspace preserves the queued task's full snapshot; fail-open
control completes the owner_fallback merge.

Docs: migration 172 comment reworded to legacy-writer/unbackfilled-lineage
semantics (source NULL is not strictly pre-migration); PR description synced.

Migration renumber vs latest main (must-fix 2) deferred to pre-launch per Bohan.

Co-authored-by: multica-agent <github@multica.ai>

* chore(attribution): renumber migrations to 167-173 after syncing main

Merged latest main (which took 166_project_dates) and shifted the attribution
migration series off the 166 collision: 166->167 agent_task_attribution,
167->168 accountable_user, 168->169 rule_version, 169->170 rule_version_index,
170->171 fail_closed, 171->172 trigger_publisher, 172->173 invariant_check.
Updated the internal cross-references in the migration comments accordingly.
Fixes TestMigrationNumericPrefixesStayUniqueAfterLegacySet on the merge tree.

Co-authored-by: multica-agent <github@multica.ai>

* feat(admission): unify dispatch outcome + close rerun/chat/autopilot invoke holes (MUL-4525)

P0 first increment toward a platform-wide execution-admission contract so a
user who names an execution target always gets a definite result and never a
silent no-op, and so blocked targets are reported without leaking private-agent
details.

Backend:
- New shared contract (handler/admission.go): DispatchOutcome / DispatchStatus
  (queued/coalesced/deferred/blocked) + stable, enumeration-safe
  DispatchReasonCode set, plus writeDispatchBlocked() whose legacy `error`
  string never reveals target existence.
- Rerun (task.go / task_lifecycle.go): re-validate the operator can invoke the
  RESOLVED target agent (historical agent for a task_id rerun) before any
  cancel/enqueue; blocked returns a structured 403 and mutates nothing
  (ErrRerunInvokeNotAllowed).
- Chat send (chat.go): re-run canInvokeAgent on every send, not just the softer
  canAccessPrivateAgent view gate; a revoked permission blocks before the
  message/attachments/task persist.
- Autopilot manual "run now" (service/autopilot.go): admission now keys on the
  current CLICKER, not the autopilot creator — clicker admission and clicker
  attribution no longer fork. Automation (schedule/webhook) still falls back to
  the creator gate. Added reason_code to the run response for the UI.

Frontend:
- triggerAutopilot response is schema-parsed; handleRunNow branches on run
  status and shows a localized, reason_code-based warning for skipped/failed
  instead of a false-success toast.
- Chat send and rerun surface the structured 403 reason_code as localized
  toasts (dispatchReasonCode helper) instead of a generic failure.
- Additive fields only; older clients keep working. i18n added to all 4 locales.

Tests: rerun fail-before-mutation gate, autopilot clicker-vs-creator fork
(service + handler), reason-code classification, and a malformed-response
schema test. Backend handler+service suites and FE typecheck/lint green.

Co-authored-by: multica-agent <github@multica.ai>

* test(cmd/server): thread nil invoke gate through RerunIssue call sites (MUL-4525)

Co-authored-by: multica-agent <github@multica.ai>

* fix(admission): typed reason codes + run-now whitelist + real security tests (MUL-4525)

Addresses Elon's review of the P0 first increment (must-fix 1–3).

1. Run now no longer has a false-success branch. handleRunNow now classifies on
   a whitelist via a pure runNowToastKind(): only issue_created/running →
   success; skipped → warning; failed and any unknown/future status → error. Add
   run-now-toast unit test over all five status classes plus reason-code → key
   mapping.

2. reason_code is a typed value decided at the admission source, not
   reverse-engineered from English failure text. New leaf package
   internal/dispatch holds the canonical ReasonCode enum (shared by handler +
   service so they can never drift). shouldSkipDispatch / errDispatchSkipped /
   the fail path now carry a typed code through DispatchAutopilotManual straight
   into the response; the substring classifier is deleted. Fixes the two missed
   branches: attribution fail-closed → attribution_blocked (typed errors.Is),
   "agent has no runtime bound" → runtime_offline. Regression tests for both.

3. Security acceptance tests exercise the REAL handlers, not injected callbacks:
   - Chat: create session while invokable → revoke invoke (flip to private, keep
     owner-view) → send returns 403 + reason_code with zero chat_message / task
     writes.
   - Rerun: private historical agent through RerunIssue + canInvokeAgent — a
     non-invoking workspace owner is refused 403 + reason_code and mutates
     nothing (fail-before-mutation); the agent owner is allowed 202.

No migration; reason_code is a decision-time value only the manual "run now"
response carries. Additive on the wire. Backend build/vet/handler+service
suites and FE typecheck/lint/vitest green. (Migration-prefix collision with main
remains the deferred pre-merge renumber.)

Co-authored-by: multica-agent <github@multica.ai>

* test(admission): make must-fix 3 acceptance tests prove the invariant (MUL-4525)

Addresses Elon's round-3 narrow review — the two security tests were not yet
falsifiable against the bugs they must catch.

- Rerun: after enqueuing the historical task, reassign the issue to a SECOND
  agent that the denied user CAN invoke. Now the current assignee and the
  task_id agent differ, so a rerun that wrongly validated the current assignee
  would let the denied user through — the 403 proves the gate is keyed on the
  historical private agent. The allow case now asserts the reran task's agent_id
  is the historical agent, not the current assignee.

- Chat: the blocked send now carries a valid, still-unbound attachment. After
  the 403 the test asserts the attachment's chat_session_id and chat_message_id
  are both still NULL, guarding against anyone moving attachment binding ahead
  of the invoke gate.

Test-only. Full internal/handler Go suite green.

Co-authored-by: multica-agent <github@multica.ai>

* feat(comments): surface blocked @mention trigger_outcomes instead of silent no-op (MUL-4525 §2)

A comment that @mentions an agent/squad the author cannot invoke used to save
with zero feedback — the user assumed a bug. Now the explicit-mention path
reports a per-target outcome on both preview and create/edit.

Backend (server/internal/handler/comment.go):
- resolveMentionedAgentCommentTriggers collects blocked outcomes instead of a
  silent `continue`. The invoke gate is evaluated BEFORE any archived/runtime
  state is read, so a caller who cannot invoke a private target only ever sees
  the generic invocation_not_allowed and can never enumerate its existence.
- enqueueCommentAgentTriggers returns queued/coalesced/deferred/blocked per
  explicit mention; enqueue errors are typed (attribution_blocked via
  errors.Is), not swallowed. Implicit routing (assignee/thread/conversation)
  carries no outcome — the user never named those targets.
- trigger-preview returns `blocked[]`; create/edit return additive
  `trigger_outcomes[]`. One blocked mention never fails the comment.

Frontend:
- Composer shows a warning chip for blocked mentions before sending; after
  sending, a "posted, but N not triggered" toast (blocked-only; coalesced/
  deferred are success-shaped). Additive schema + defensive parse; i18n ×4.

Tests: handler partial-success + enumeration-safety acceptance tests; core
outcome-parse + preview-schema tests; hook/parity updated. Backend
handler+service suites and FE typecheck/lint/vitest green.

Co-authored-by: multica-agent <github@multica.ai>

* i18n(admission): clearer, consistent blocked-trigger copy (MUL-4525)

Reword the awkward "you are not allowed to run this autopilot's assignee" and
polish all MUL-4525 blocked/partial copy across en/zh-Hans/ja/ko:

- Unify on "you don't have permission to use this <agent|target>" (zh: 没有…的
  使用权限) for autopilot Run now, comment mentions, rerun, and chat send —
  replacing the various "not allowed to run/trigger" phrasings.
- Align zh to the glossary term 智能体 (was mixed "Agent"), matching the
  surrounding UI voice (e.g. agent_link_no_access).
- Drop jargon: "blocked by an admission policy"/"被准入策略拦截" → "the run was
  blocked"/"本次运行已被拦截"; "attributed"/"归因" → plainer wording.
- Comment copy counts "mentions" (was "targets") for consistency with the chip.

Backend dispatchBlockedFallbackMessage (old-client English fallback) reworded to
match; its enumeration-safety test assertion updated. Copy-only — no key/logic
changes; parity + typecheck green.

Co-authored-by: multica-agent <github@multica.ai>

* fix(comments): one outcome per explicit mention + FE success whitelist (MUL-4525)

Addresses Elon's round-2 review of the §2 comment trigger_outcomes.

1. Separate execution dedup from per-target outcomes. resolveMentionedAgent-
   CommentTriggers now returns triggers (deduped by executing agent) AND one
   commentMentionTarget per EXPLICIT mention. enqueueCommentAgentTriggers returns
   a per-executing-agent result map; commentTriggerOutcomes fans each agent's
   status to every target that resolved to it. So @Agent A + @Squad S(leader=A)
   coalesces to ONE task but yields TWO outcomes. The squad-leader self-suppress
   branch now returns a definite `deferred` outcome instead of no result. New
   CreateComment test asserts 1 task, 2 outcomes.

2. Frontend no longer treats an unknown status as success. unhandledComment-
   TriggerOutcomes whitelists queued/coalesced/deferred as handled; blocked and
   any unknown/future/empty status warn (mirrors the Run now whitelist). The
   preview schema's `blocked` now drops malformed entries INDIVIDUALLY instead of
   z.array(...).catch([]) discarding the whole set. Regression tests for the
   unknown status and the per-item drop.

Backend handler suite + go vet, FE typecheck/lint/vitest green.

Co-authored-by: multica-agent <github@multica.ai>

* fix(comments): honest role + status in trigger_outcomes fan-out (MUL-4525)

Addresses Elon's round-3 review of the §2 fan-out.

1. Execution merge preserves the squad-leader role instead of first-mention-
   wins. The dedup now UPGRADES an already-added plain @agent trigger to a
   @squad-leader trigger for the same agent, so @Agent A + @Squad S(leader=A)
   always runs as a leader task (is_leader_task + squad_id=S) regardless of
   mention order — the daemon still injects S's briefing. When two DIFFERENT
   squads share one leader, the single run carries one squad's context and the
   other squad is reported `coalesced` (folded), never a second `queued`. The
   enqueue result now records the executed squad so the fan-out can tell them
   apart. Tests assert the task role in both orders and the two-squad split.

2. Squad-leader self-suppression no longer fakes success. The self-trigger
   guard keys on the latest task ROLE with no status filter, so a long-completed
   task also suppresses; reporting `deferred/already_active` when nothing is
   active was a false success. The branch now reports `deferred` only when a
   real non-terminal task is active (its reconcile covers the comment), else a
   non-success `blocked` + new `already_handled` reason. Fixed the reversed
   helper-semantics comment. New handler test covers the completed-task branch.

Backend handler+service suites and go vet green.

Co-authored-by: multica-agent <github@multica.ai>

* fix(comments): fail-closed active-task check, never fake deferred (MUL-4525)

Elon round-4 must-fix: hasActiveTaskForIssueAndAgent swallowed DB errors by
returning true, so both call sites could turn a query failure into a
success-shaped `deferred/already_active` — a silent false success, exactly what
this issue forbids for admission-query failures.

- hasActiveTaskForIssueAndAgent now returns (bool, error).
- Two pure decision helpers govern the branches, fail closed on error:
  - decidePostMergeMiss: on query error, do NOT enqueue a fresh task (duplicate
    concurrent-run risk) AND report non-success blocked/internal_error; a
    confirmed active task defers; a confirmed-none enqueues fresh.
  - decideSuppressedLeaderOutcome: on query error, blocked/internal_error; a
    confirmed active run defers; else self_trigger_suppressed. Never a fabricated
    deferred.
- Renamed reason already_handled -> self_trigger_suppressed (Elon non-blocking
  note): the old name implied the new comment was already processed, but the
  real meaning is a suppressed self-trigger.

Deterministic unit tests cover the query-failure branch at both call sites
(no fresh enqueue, non-success outcome) — a real DB fault can't be forced
through valid handler inputs, and the decision is what governs the behavior.
Backend handler+service suites and go vet green.

Co-authored-by: multica-agent <github@multica.ai>

* fix(comments): honest merge outcome — refused merge is blocked, not fake coalesced (MUL-4525)

A pending-task merge previously reported success (coalesced) even when it
was refused or failed: attribution fail-closed and unknown DB errors both
returned handled=true, so the caller recorded coalesced for a merge that
never happened. mergeCommentIntoPendingTask now returns a distinguishable
commentMergeResult; commentMergeTerminalOutcome maps a real merge to
coalesced, a fail-closed refusal to blocked/attribution_blocked, and any
other failure to blocked/internal_error. Only "no queued task to fold"
falls through to the active-task decision. Adds pure coverage of the
mapping plus a fail-closed regression asserting the non-success outcome
and unchanged task count.

Co-authored-by: multica-agent <github@multica.ai>

* fix(comments): name blocked @mentions in preview + toast instead of a vague count (MUL-4525)

The blocked-trigger preview showed a red "1 mention won't trigger" with no
name, and the post-send toast said "but 1 mention wasn't triggered" — the
user can't tell which target or why. Now each blocked mention renders its own
chip named from the mention markup the user typed ("Go · No permission"),
with an error indicator and a short reason; the toast names the single target
too. The wire outcome still omits the target name (enumeration-safety) — the
label comes from the user's own draft, so nothing new is disclosed.

Shares a blocked-trigger-copy module (long + short reason labels) between the
chip and the toast, and a pure mentionLabelsByTarget/parseMentions helper in
core (fresh regex per call — a shared global leaked lastIndex). Adds core +
chip tests; drops the now-unused trigger_blocked_count keys across locales.

Co-authored-by: multica-agent <github@multica.ai>

* feat(attribution): accountable-member avatar on agent task rows + transcript header (MUL-4302)

Surface who each agent run is on behalf of where runs are actually
browsed:
- Agent detail activity tab: an avatar-only AttributionBadge on every
  task row's meta line (Now + Recent work), tooltip carries the name +
  resolution source.
- Execution-record (transcript) dialog header: the full on-behalf-of
  badge next to the status pill.

Adds a compact variant="avatar" mode to AttributionBadge, reusing its
source-label mapping and degraded-attribution tone. Renders nothing when
a run has no resolved accountable member.

Co-authored-by: multica-agent <github@multica.ai>

---------

Co-authored-by: J <j@multica.ai>
Co-authored-by: multica-agent <github@multica.ai>
2026-07-15 03:09:56 +08:00

545 lines
21 KiB
SQL

-- =====================
-- Autopilot CRUD
-- =====================
-- name: ListAutopilots :many
-- List rows carry three derived columns the list UI needs (trigger badges,
-- next run, last-run outcome) so the page never has to N+1 into the detail
-- endpoint. trigger_kinds/next_run_at only consider ENABLED triggers — the
-- columns answer "how does this fire today", not "what is configured".
-- last_run_status is COALESCEd to '' (never ran) because sqlc cannot infer
-- nullability through a scalar subquery; the handler maps '' back to omitted.
SELECT
sqlc.embed(a),
(
SELECT array_agg(DISTINCT t.kind ORDER BY t.kind)
FROM autopilot_trigger t
WHERE t.autopilot_id = a.id AND t.enabled
)::text[] AS trigger_kinds,
(
SELECT min(t.next_run_at)
FROM autopilot_trigger t
WHERE t.autopilot_id = a.id AND t.enabled AND t.kind = 'schedule'
)::timestamptz AS next_run_at,
COALESCE((
SELECT r.status
FROM autopilot_run r
WHERE r.autopilot_id = a.id
ORDER BY r.triggered_at DESC
LIMIT 1
), '')::text AS last_run_status
FROM autopilot a
WHERE a.workspace_id = $1
AND (
(sqlc.narg('status')::text IS NULL AND a.status <> 'archived')
OR a.status = sqlc.narg('status')
)
ORDER BY a.created_at DESC;
-- name: GetAutopilot :one
SELECT * FROM autopilot
WHERE id = $1;
-- name: GetAutopilotInWorkspace :one
SELECT * FROM autopilot
WHERE id = $1 AND workspace_id = $2;
-- name: CreateAutopilot :one
INSERT INTO autopilot (
workspace_id, title, description, assignee_type, assignee_id,
status, execution_mode, issue_title_template, project_id,
created_by_type, created_by_id
) VALUES (
$1, $2, sqlc.narg('description'), $3, $4,
$5, $6, sqlc.narg('issue_title_template'), sqlc.narg('project_id'),
$7, $8
) RETURNING *;
-- name: UpdateAutopilot :one
UPDATE autopilot SET
title = COALESCE(sqlc.narg('title'), title),
description = COALESCE(sqlc.narg('description'), description),
assignee_type = COALESCE(sqlc.narg('assignee_type'), assignee_type),
assignee_id = COALESCE(sqlc.narg('assignee_id')::uuid, assignee_id),
status = COALESCE(sqlc.narg('status'), status),
execution_mode = COALESCE(sqlc.narg('execution_mode'), execution_mode),
issue_title_template = sqlc.narg('issue_title_template'),
project_id = sqlc.narg('project_id'),
updated_at = now()
WHERE id = $1
RETURNING *;
-- name: ArchiveAutopilot :exec
UPDATE autopilot
SET status = 'archived', updated_at = now()
WHERE id = $1;
-- name: UpdateAutopilotLastRunAt :exec
UPDATE autopilot SET last_run_at = now(), updated_at = now()
WHERE id = $1;
-- =====================
-- Autopilot Rule Version (rule_owner attribution, MUL-4302 §3.4)
-- =====================
-- name: CreateAutopilotRuleVersion :one
-- Append one immutable rule-version snapshot on a substantive publish (create /
-- enable / resume / target / execution-mode change). published_by_* is the acting
-- member (or 'system' with NULL id for the failure monitor); config_summary is the
-- effective config at publish time. Dispatch reads the latest row for the autopilot
-- as the run's rule_owner accountable human.
INSERT INTO autopilot_rule_version (
autopilot_id, workspace_id, published_by_type, published_by_id, config_summary
)
VALUES (
@autopilot_id, @workspace_id, @published_by_type, sqlc.narg(published_by_id),
COALESCE(sqlc.narg(config_summary), '{}'::jsonb)
)
RETURNING *;
-- name: GetActiveAutopilotRuleVersion :one
-- The active version is the newest published row for the autopilot. Scoped by
-- workspace_id per the workspace query rule; autopilot_id is globally unique so the
-- workspace filter is a guard, not the selector.
SELECT * FROM autopilot_rule_version
WHERE workspace_id = $1 AND autopilot_id = $2
ORDER BY created_at DESC
LIMIT 1;
-- =====================
-- Autopilot Trigger CRUD
-- =====================
-- name: ListAutopilotTriggers :many
SELECT * FROM autopilot_trigger
WHERE autopilot_id = $1
ORDER BY created_at ASC;
-- name: GetAutopilotTrigger :one
SELECT * FROM autopilot_trigger
WHERE id = $1;
-- name: CreateAutopilotTrigger :one
INSERT INTO autopilot_trigger (
autopilot_id, kind, enabled, cron_expression, timezone,
next_run_at, webhook_token, label, provider, event_filters,
published_by_type, published_by_id
) VALUES (
$1, $2, $3, sqlc.narg('cron_expression'), sqlc.narg('timezone'),
sqlc.narg('next_run_at'), sqlc.narg('webhook_token'), sqlc.narg('label'),
COALESCE(sqlc.narg('provider')::text, 'generic'),
sqlc.narg('event_filters'),
sqlc.narg('published_by_type'), sqlc.narg('published_by_id')
) RETURNING *;
-- name: SetAutopilotTriggerPublisher :exec
-- Re-stamp a single trigger's responsible publisher after a substantive edit of
-- THAT trigger (cron / filter / enabled / webhook security). Future runs it fires
-- become accountable to this member (MUL-4302 trigger_owner transfer).
UPDATE autopilot_trigger
SET published_by_type = $2, published_by_id = $3, updated_at = now()
WHERE id = $1;
-- name: SetAutopilotTriggerPublishersByAutopilot :exec
-- Re-stamp ALL of an autopilot's triggers' responsible publisher after a substantive
-- AUTOPILOT-level edit (target / instructions / assignee / execution-mode / enable).
-- Such a change governs every trigger's future runs, so responsibility transfers to
-- the editing member for all of them; a per-trigger edit uses the single-trigger
-- variant so it never reassigns another trigger (MUL-4302).
UPDATE autopilot_trigger
SET published_by_type = $2, published_by_id = $3, updated_at = now()
WHERE autopilot_id = $1;
-- name: UpdateAutopilotTrigger :one
UPDATE autopilot_trigger SET
enabled = COALESCE(sqlc.narg('enabled')::boolean, enabled),
cron_expression = COALESCE(sqlc.narg('cron_expression'), cron_expression),
timezone = COALESCE(sqlc.narg('timezone'), timezone),
next_run_at = sqlc.narg('next_run_at'),
label = COALESCE(sqlc.narg('label'), label),
event_filters = COALESCE(sqlc.narg('event_filters'), event_filters),
updated_at = now()
WHERE id = $1
RETURNING *;
-- name: DeleteAutopilotTrigger :exec
DELETE FROM autopilot_trigger WHERE id = $1;
-- name: AdvanceTriggerNextRun :exec
UPDATE autopilot_trigger
SET next_run_at = sqlc.narg('next_run_at'),
last_fired_at = now(),
updated_at = now()
WHERE id = $1;
-- name: GetWebhookTriggerByToken :one
-- Look up a webhook trigger by its public bearer token. Joined to autopilot
-- so the webhook handler can derive the workspace from the trigger's parent
-- without trusting any request header. The handler still re-loads the
-- Autopilot via GetAutopilot and cross-checks WorkspaceID matches the row's
-- autopilot_workspace_id.
SELECT t.*, a.workspace_id AS autopilot_workspace_id
FROM autopilot_trigger t
JOIN autopilot a ON a.id = t.autopilot_id
WHERE t.kind = 'webhook'
AND t.webhook_token = $1;
-- name: TouchAutopilotTriggerFiredAt :exec
-- Bumps last_fired_at after a webhook fires, regardless of whether the
-- dispatch succeeded, was admission-skipped, or even if Autopilot status
-- transitioned to paused/disabled at exactly the wrong moment. Disabled /
-- paused early-return paths in the handler never call this.
UPDATE autopilot_trigger
SET last_fired_at = now(),
updated_at = now()
WHERE id = $1;
-- name: RotateAutopilotTriggerWebhookToken :one
-- Rotates the bearer token for a webhook trigger. Restricted to kind='webhook'
-- so an accidental call against a schedule/api trigger is a no-op (returns no
-- rows) rather than corrupting unrelated state.
UPDATE autopilot_trigger
SET webhook_token = $2,
updated_at = now()
WHERE id = $1
AND kind = 'webhook'
RETURNING *;
-- name: SetAutopilotTriggerWebhookToken :one
-- Sets the webhook token at creation time. CreateAutopilotTrigger inserts the
-- row first (using its full 8-arg signature), then this query attaches the
-- token. Splitting the create + token-set keeps the existing CreateAutopilotTrigger
-- query usable by the schedule path without forcing every caller to think
-- about webhook_token.
UPDATE autopilot_trigger
SET webhook_token = $2,
updated_at = now()
WHERE id = $1
RETURNING *;
-- name: SetAutopilotTriggerSigningSecret :one
-- Writes the signing secret for a webhook trigger. Kept as a dedicated query
-- (not a field on UpdateAutopilotTrigger) so the request body for the
-- write-only endpoint only ever carries the secret value, with no risk of an
-- accidental log line leaking it alongside other fields. Restricted to
-- webhook triggers to avoid corrupting unrelated state.
UPDATE autopilot_trigger
SET signing_secret = sqlc.narg('signing_secret'),
updated_at = now()
WHERE id = $1
AND kind = 'webhook'
RETURNING *;
-- =====================
-- Autopilot Run Management
-- =====================
-- name: CreateAutopilotRun :one
-- squad_id is an attribution hook: set to the assignee squad when the
-- parent autopilot has assignee_type='squad', NULL otherwise. The executing
-- agent_id on agent_task_queue still records who actually ran the work
-- (the squad leader); squad_id lets reports group by squad without a join.
--
-- planned_at carries the canonical UTC fire time for scheduled triggers
-- (source='schedule'); it stays NULL for manual / webhook / api sources
-- which have no canonical occurrence. Combined with the partial unique
-- index uq_autopilot_run_trigger_planned, this gives dispatch-layer
-- idempotency: a stale-steal retry at the same plan_time cannot create
-- a second run for the same (trigger_id, planned_at) pair (MUL-3551).
INSERT INTO autopilot_run (
autopilot_id, trigger_id, source, status, trigger_payload, squad_id, planned_at,
webhook_delivery_id
) VALUES (
$1, sqlc.narg('trigger_id'), $2, $3, sqlc.narg('trigger_payload'),
sqlc.narg('squad_id'), sqlc.narg('planned_at'),
sqlc.narg('webhook_delivery_id')
) RETURNING *;
-- name: GetAutopilotRunByTriggerAndPlanned :one
-- Idempotent lookup used by DispatchAutopilotForPlan to detect a
-- crash-during-dispatch retry: if a row already exists for this
-- (trigger_id, planned_at), the caller reuses it instead of creating a
-- duplicate. The partial unique index covers the same key, so a race
-- between "look up then insert" still resolves to a single row — this
-- query is just the fast path that lets us skip the INSERT when we
-- can see the prior row clearly. Returns no rows for the (much more
-- common) first-time dispatch.
SELECT * FROM autopilot_run
WHERE trigger_id = $1
AND planned_at = $2
LIMIT 1;
-- name: GetAutopilotRunByWebhookDelivery :one
SELECT * FROM autopilot_run
WHERE webhook_delivery_id = $1
LIMIT 1;
-- name: RecoverPartialAutopilotRun :exec
-- Recovers a partial-state autopilot_run from a crashed first attempt
-- (the runner wrote the run row but died before creating the downstream
-- issue/task) so that a subsequent DispatchAutopilotForPlan call can
-- create a fresh run at the same (trigger_id, planned_at).
--
-- Setting planned_at = NULL clears the partial-unique slot held by
-- uq_autopilot_run_trigger_planned, letting the new INSERT proceed.
-- The row stays in autopilot_run as a FAILED record (with a recovery
-- reason) so ops still see the abandoned attempt in the run history —
-- it is not silently deleted.
UPDATE autopilot_run
SET status = 'failed',
completed_at = now(),
failure_reason = 'recovered partial dispatch (crashed before downstream creation)',
planned_at = NULL
WHERE id = $1;
-- name: GetAutopilotRun :one
SELECT * FROM autopilot_run
WHERE id = $1;
-- name: ListAutopilotRuns :many
SELECT * FROM autopilot_run
WHERE autopilot_id = $1
ORDER BY created_at DESC
LIMIT $2 OFFSET $3;
-- name: UpdateAutopilotRunIssueCreated :one
UPDATE autopilot_run
SET status = 'issue_created', issue_id = $2
WHERE id = $1
RETURNING *;
-- name: UpdateAutopilotRunRunning :one
UPDATE autopilot_run
SET status = 'running', task_id = $2
WHERE id = $1
RETURNING *;
-- name: UpdateAutopilotRunCompleted :one
UPDATE autopilot_run
SET status = 'completed', completed_at = now(), result = sqlc.narg('result')
WHERE id = $1
RETURNING *;
-- name: UpdateAutopilotRunFailed :one
UPDATE autopilot_run
SET status = 'failed', completed_at = now(), failure_reason = $2
WHERE id = $1
RETURNING *;
-- name: UpdateAutopilotRunSkipped :one
-- Marks an autopilot_run as skipped without enqueueing any task. Used by the
-- pre-flight admission check when the assignee agent's runtime is offline:
-- creating an issue / task in that state would just pile a doomed job onto
-- agent_task_queue (the canonical "持续给离线 local agent 入队" symptom from
-- MUL-1899). Recording the skip + reason gives the UI / failure monitor / ops
-- a paper trail without polluting the failure ratio.
UPDATE autopilot_run
SET status = 'skipped', completed_at = now(), failure_reason = $2
WHERE id = $1
RETURNING *;
-- name: UpdateAutopilotRunSkippedWithResult :one
UPDATE autopilot_run
SET status = 'skipped',
completed_at = now(),
failure_reason = $2,
result = sqlc.narg('result')
WHERE id = $1
RETURNING *;
-- =====================
-- Scheduler Queries
-- =====================
-- name: ListSchedulableAutopilotTriggers :many
-- Lists every schedule trigger the autopilot_schedule_dispatch JobSpec
-- should consider this tick. Returns just the columns the scheduler's
-- scope provider + PlansForScope hook need; the full trigger row is
-- re-loaded by the handler so a trigger update between scope-list and
-- handler-run sees the latest enabled / cron values.
--
-- last_fired_at is read so the planner hook can anchor cold-start
-- enumeration on the most recent successful fire (set by either the
-- legacy goroutine before the new scheduler took over, or the new
-- scheduler's own post-dispatch advance — AdvanceTriggerNextRun, falling
-- back to TouchAutopilotTriggerFiredAt on a cron parse error). Without it,
-- a trigger that was created days ago and fired by the legacy code
-- looks like a brand-new trigger to the new scheduler on first tick
-- and the half-open `(created_at, now]` enumeration replays the most
-- recent already-fired occurrence — exactly the post-deploy
-- spurious-fire reported on MUL-3551 dev.
--
-- Filters out webhook / api triggers, disabled triggers, paused/archived
-- autopilots, and any trigger missing its cron expression. ORDER BY id
-- keeps the per-tick scope list stable across replicas.
SELECT t.id, t.autopilot_id, t.cron_expression, t.timezone, t.created_at, t.last_fired_at
FROM autopilot_trigger t
JOIN autopilot a ON a.id = t.autopilot_id
WHERE t.kind = 'schedule'
AND t.enabled = TRUE
AND a.status = 'active'
AND t.cron_expression IS NOT NULL
AND t.cron_expression <> ''
ORDER BY t.id;
-- =====================
-- Task Queue (run_only mode)
-- =====================
-- name: CreateAutopilotTask :one
-- run_only autopilot dispatch. Attribution depends on the trigger:
-- * schedule / webhook / api: no human authorized the run, so originator_user_id
-- stays NULL and accountable_user_id is the rule_owner (the publisher of the
-- autopilot's active rule version), with rule_version_id recording the snapshot
-- (MUL-4302 §3.4) — the accountable-diverges-from-originator case.
-- * manual: a member clicked "run now", a direct human action, so originator and
-- accountable are BOTH that member (originator_source='direct_human'); no rule
-- version is involved (MUL-4302 §4).
-- When no version/publisher resolves on the non-manual path, the caller passes NULL
-- accountable + originator_source='unattributed' so the row is still not a
-- NULL-source bypass (MUL-4302 §2).
INSERT INTO agent_task_queue (
agent_id, runtime_id, issue_id, status, priority, autopilot_run_id, trigger_summary,
originator_user_id, accountable_user_id, rule_version_id,
originator_source, trigger_evidence_kind, trigger_evidence_ref_id
)
VALUES (
$1, $2, NULL, 'queued', $3, $4, sqlc.narg(trigger_summary),
sqlc.narg(originator_user_id),
sqlc.narg(accountable_user_id),
sqlc.narg(rule_version_id),
sqlc.narg(originator_source),
sqlc.narg(trigger_evidence_kind),
sqlc.narg(trigger_evidence_ref_id)
)
RETURNING *;
-- name: GetAutopilotTaskByRun :one
-- Repairs the narrow run_only crash window where the task INSERT committed
-- but the following autopilot_run.task_id update did not.
SELECT * FROM agent_task_queue
WHERE autopilot_run_id = $1
ORDER BY created_at
LIMIT 1;
-- =====================
-- Run lookup by linked entities
-- =====================
-- name: GetAutopilotRunByIssue :one
SELECT * FROM autopilot_run
WHERE issue_id = $1 AND status IN ('issue_created', 'running')
LIMIT 1;
-- name: FailAutopilotRunsByIssue :exec
-- Fails active autopilot runs linked to a given issue.
-- Must be called BEFORE issue deletion (ON DELETE SET NULL clears issue_id).
UPDATE autopilot_run
SET status = 'failed', completed_at = now(), failure_reason = 'linked issue was deleted'
WHERE issue_id = $1
AND status IN ('issue_created', 'running');
-- =====================
-- Failure-rate auto-pause
-- =====================
-- name: SelectAutopilotsExceedingFailureThreshold :many
-- Find active autopilots whose recent run failure rate exceeds the threshold.
-- Counts only "real" terminal runs (completed | failed). 'skipped' is
-- excluded from BOTH numerator and denominator: an admission-skipped run
-- (e.g. assignee runtime offline at dispatch time, MUL-1899) is neither a
-- success nor a failure, so it must not dilute the failure ratio (which
-- would let a 100%-failing autopilot mask itself behind a wall of skips)
-- nor inflate it. issue_created/running are still excluded so in-flight
-- work isn't penalised.
-- Used by the failure monitor to auto-pause sustained-failure autopilots
-- (the canonical example from MUL-1336 was an autopilot scheduled every 5 min
-- that 100% failed for days, burning ~1.5k useless tasks per week).
WITH stats AS (
SELECT autopilot_id,
count(*) FILTER (WHERE status IN ('completed', 'failed')) AS total,
count(*) FILTER (WHERE status = 'failed') AS failed
FROM autopilot_run
WHERE created_at >= sqlc.arg('since')::timestamptz
GROUP BY autopilot_id
)
SELECT a.id, a.workspace_id, a.title, a.assignee_id,
a.created_by_type, a.created_by_id,
s.total::bigint AS total_runs,
s.failed::bigint AS failed_runs
FROM autopilot a
JOIN stats s ON s.autopilot_id = a.id
WHERE a.status = 'active'
AND s.total >= sqlc.arg('min_runs')::bigint
AND s.failed::float8 / NULLIF(s.total, 0)::float8 >= sqlc.arg('fail_ratio_threshold')::float8
ORDER BY s.failed DESC, a.id ASC;
-- name: SystemPauseAutopilot :one
-- Atomically pauses an autopilot only if it is currently active. Returns no
-- rows when the autopilot was already paused/archived (or another worker
-- raced first), letting the caller treat that as a benign no-op rather than
-- an error.
UPDATE autopilot
SET status = 'paused', updated_at = now()
WHERE id = $1 AND status = 'active'
RETURNING *;
-- =====================
-- Autopilot Subscribers
-- =====================
-- name: ListAutopilotSubscribers :many
-- ORDER BY created_at keeps chip rendering stable across refreshes.
SELECT * FROM autopilot_subscriber
WHERE autopilot_id = $1
ORDER BY created_at ASC, user_id ASC;
-- name: AddAutopilotSubscriber :exec
INSERT INTO autopilot_subscriber (autopilot_id, user_type, user_id)
VALUES ($1, $2, $3)
ON CONFLICT (autopilot_id, user_type, user_id) DO NOTHING;
-- name: DeleteAutopilotSubscribersForAutopilot :exec
-- Paired with a re-insert loop to implement full-replace PATCH semantics.
DELETE FROM autopilot_subscriber
WHERE autopilot_id = $1;
-- =====================
-- Autopilot Collaborators
-- =====================
-- name: ListAutopilotCollaborators :many
-- ORDER BY created_at keeps row rendering stable across refreshes.
SELECT * FROM autopilot_collaborator
WHERE autopilot_id = $1
ORDER BY created_at ASC, user_id ASC;
-- name: AddAutopilotCollaborator :one
-- Re-granting an existing collaborator is a no-op that refreshes granted_by,
-- so the call is idempotent from the API boundary.
INSERT INTO autopilot_collaborator (autopilot_id, user_type, user_id, granted_by)
VALUES ($1, $2, $3, $4)
ON CONFLICT (autopilot_id, user_type, user_id)
DO UPDATE SET granted_by = EXCLUDED.granted_by
RETURNING *;
-- name: DeleteAutopilotCollaborator :exec
DELETE FROM autopilot_collaborator
WHERE autopilot_id = $1 AND user_type = $2 AND user_id = $3;
-- name: DeleteAutopilotCollaboratorsForAutopilot :exec
-- Application-layer cleanup run inside the autopilot delete transaction.
DELETE FROM autopilot_collaborator
WHERE autopilot_id = $1;
-- name: IsAutopilotCollaborator :one
SELECT EXISTS (
SELECT 1 FROM autopilot_collaborator
WHERE autopilot_id = $1 AND user_type = 'member' AND user_id = $2
) AS is_collaborator;
-- name: ListAutopilotIDsForCollaborator :many
-- Powers the per-row can_write flag on the list endpoint without an N+1.
SELECT autopilot_id FROM autopilot_collaborator
WHERE user_type = 'member' AND user_id = $1;