Commit Graph
3 Commits
Author SHA1 Message Date
48556a032b Fix: CEO reject conduit and main pm routing (#91)
* fix(tasks): deliver CEO change-requests to the reworker and route integration rejects to the Main PM

ceo_reject previously appended the CEO's required-changes only to quick_context
(never surfaced to an agent) and always reassigned to the original developer, so
the feedback never reached whoever reworked the task and a coordination/integration
root went to a developer instead of the Main PM.

Now ceo_reject records the reason as a DECISION_LOG handoff journal entry — the
channel evidence/journal_highlights already serves to the task's assignee — and
routes a coordination task (no project, has product) to team=main_pm + the Main PM
so it can delegate the rework. Leaf dev tasks still return to the original
developer. The journal write is best-effort (author-exists guard) and never blocks
a reject. Adds tests for the routing and the handoff-journal write.

* feat(gateway): wire the agent context_briefing receive-path

EvidenceRepo's agent-scoped methods were stubbed to return empty lists, so agents
never received notifications, A2A DMs, @mentions, recent team activity, or in-lane
blockers through their briefing — the system was effectively send-only. This
implements all five as single capped queries over the live tables (plus a light
task-metadata-gap check), mirroring the existing journal_highlights query. Each runs
on the per-verb briefing path, so each stays a single LIMIT-10 indexed lookup.

Unit tests cover the row mapping + empty paths; an integration suite exercises the
real SQL (array contains, the a2a slug filter, team+status) against Postgres.

---------

Co-authored-by: Renn F <rennf93@users.noreply.github.com>
2026-06-10 02:26:45 +02:00
Renn F a5af6159d7 fix(gateway): evidence surfaces the task's board/PM handoff journal entries
journal_highlights_for_task was a Phase-1 stub returning [], so evidence()
never surfaced the upstream Product Owner / Head of Marketing review. The
Main PM picked up a board-reviewed coordination task, saw no handoff, and
re-researched from scratch (duplicated work). Query the task's
decision/reflection/note entries across authors (slug + role), oldest
first, so the Main PM builds on the board's analysis instead of redoing it.
2026-06-03 08:27:10 +02:00
Renn F 9aa30fb945 100% Coverage 2026-05-06 21:02:31 +02:00