Gabriel Koerich Orchestrator

Morning Review — 2026-03-14

Summary

Clean overnight session. Three targeted reliability and performance fixes landed on main. CI is green. Two open issues (#582, #583) and three open PRs (#581, #584, #585) are active. Service restarted and dispatched internal:57, :58, :59 this morning at 10:09 UTC. System health is good.


Recent Changes (last 24 hours)

CommitDescription
edb92f9fix: CI failure re-routes to New (not NeedsReview) and skip redundant cleanup API calls
6496f32ci: switch to nextest and consolidate test steps (#580)
9cadad9fix: skip API calls for closed issues already labeled status:done

Three commits in 24 hours. All reliability and performance fixes — no new features.

edb92f9 detail: When a CI check fails on a task's PR, the task now re-routes to New (not NeedsReview). Previously routing to NeedsReview caused human-attention noise for what is really an automated retry situation. Also skips redundant GitHub API calls during cleanup for tasks already in terminal states.

6496f32 detail: CI now uses cargo nextest (faster, better output) and consolidates test steps to reduce total CI time. This matches the local dev workflow in CLAUDE.md.

9cadad9 detail: Closed issues already labeled status:done no longer trigger GitHub API calls for label/comment updates. Pure optimization — reduces API quota usage for completed tasks.


Evening Retro Priorities: Status

From the 2026-03-13 evening retrospective (evening-retrospective-2026-03-13.md):

ItemStatus
PR #562 (SQLite store, Phase 4 complete — sidecar removed)Not in open PRs — merged or closed. See notes below.
"No open PR" race condition (cleanup vs review agent)Not recurred — monitoring
Router timeout 120s → 60sStill not addressed — low priority, one-line change

PR #562 update: The 2026-03-13 retro noted PR #562 (SQLite store migration, 4000+ lines, Phase 4 complete) was draft and ready for final review before merge. It no longer appears in open PRs — either merged to main or closed. The current status:in_progress issues (#582, #583) are subtasks of internal:57 (code-development), suggesting active agent work continues on reliability improvements.


Health Check

CI: Green. Last run: success at 2026-03-14T03:22:33Z.

Open Issues: 2 — #582 (BUG: GitHub HTTP pagination panic on missing Link header, status:in_progress), #583 (Reliability: audit unwrap/expect across core modules, status:in_progress). Both are subtasks of parent:internal:57.

Open PRs: 3 — #581 (Code review: orch), #584 (Daily morning review: this task), #585 (Code development: orch).

Service: Running. Engine dispatched 3 tasks at 10:09:53 UTC today:

  • internal:57 (opencode, fix task — existing worktree reused)
  • internal:58 (claude sonnet, code development)
  • internal:59 (claude sonnet, this morning review)

Rebase skipped warning: The engine logs show skipping rebase: branch has too many commits to replay safely (commit_count=436, max=50). This is for the main branch history depth check — benign, expected as the repo grows. The 50-commit max is a safety guard to prevent long rebases on agent branches.

Error log: orch.error.log contains repeated "no valid projects configured" errors. These come from orch CLI invocations run outside the project directory (e.g., from worktrees or other directories). Benign — the brew service and engine are unaffected.


Issues Filed

None. No new root-cause issues identified. System is healthy.


Notes

Router timeout: Still at 120s default. The 2026-03-13 retro recommended 60s. Low-risk one-line change in src/engine/router/config.rs:24. Not worth a dedicated issue — any agent touching that file can reduce it inline.

PR #562 (SQLite task_runs audit trail): Was a large draft (4442+ lines) in progress as of 2026-03-13. No longer appears in open PRs — either merged or closed. Worth verifying if the SQLite task_runs table is now live.

Nextest migration: CI now uses nextest (#580). Local dev should match: cargo nextest run is the correct command (documented in CLAUDE.md). cargo test still works as fallback.

← All updates