Summary

High-throughput night: ~15 issues closed since yesterday, focused on auto-merge reliability, review parsing, and operational hardening. The auto-merge rollout continues to surface edge cases; newer issues (#1030-#1032) are now in the pipeline targeting websocket and review-retry behavior. Two critical correctness fixes landed this morning (#1027, #1028). Service running on v0.56.36, healthy.

Recent Activity (Last 24h)

Key Commits

• Stale worktree metadata auto-pruned (4b7738e) — cleanup no longer leaves orphan metadata entries after a worktree is removed
• has_workflows error propagated (#1027) — auto_merge.rs was masking lookup errors with unwrap_or(false), which could allow a PR to merge without CI running; now returns the error so the poll retries
• Review poll timestamp fix (#1028) — last_review_ts is now persisted only after handle_review_changes() succeeds, preventing silent watermark advances on transient failures
• Deterministic WS port selection (#1024) — eliminates port collision on ws server startup
• Auto-merge CI skip fix (#1016) — auto-merge was treating missing CI data as a success; now blocks
• Review envelope parsing split (#1014) — each agent format parsed independently, no more generic fallback reliance
• Duplicate review dispatch removed (#1012) — tick was dispatching reviews twice; deduplicated
• Global task listing filters (#1011) — orch task list now supports global filters
• Null CI conclusion treated as pending (#1013) — completed check runs with null conclusion no longer look like failures
• Worktree cleanup idempotency (#994) — cleanup returning Err instead of Ok(false) when no worktree/branch exists
• Review cycle cap enforced in poll (#1003) — review poll now enforces max_review_cycles
• Auto-merge closed PR idempotent (#997) — closed/merged PRs during auto-merge now return success instead of failing
• CI semaphore scope fix (#991) — semaphore no longer held across the entire auto-merge loop sleep/retry

Active Pipeline (as of now)

Operational Health

Service

• Running v0.56.36 (confirmed via log). Previous stale entries referencing 0.56.35 are leftover from the upgrade restart — expected and benign.
• orch CLI reports v0.36.1 — this is likely the brew-installed CLI being out of sync with the service binary. Worth checking with brew upgrade orch if commands behave unexpectedly.

Patterns

• Websocket reliability is the new focus: #1031 (port race) and #1032 (silent lag) follow on from the deterministic port selection fix in #1024 — further edge cases in startup and observability.
• Review retry correctness (#1030): review agent creating a missing PR is valid recovery behavior, but was being counted as a failure and consuming retry budget. Needs to be treated as a successful retry, not a failure.
• Correctness fixes for auto-merge and review poll (#1027, #1028) landed this morning — these close two subtle data-loss/skip bugs that were present since the auto-merge feature shipped.

No stuck or blocked tasks

All tasks are actively in progress, routed, or awaiting review agent pickup. No tasks are stuck or require human intervention.

Retrospective Follow-ups (from 2026-03-25 evening)

• Fix auto-merge follow-ups (#991, #997, #1003, #1016, #1027 all landed)
• Tighten retry/review parsing paths (#1014 split parsing by agent format)
• Review poll timestamp correctness (#1028 landed)
• Worktree recovery hardening (#1021 needs_review — worktree cleanup idempotency fix in review)
• Push-retry status mapping (#992 landed routed→new fix; #981 original — confirm fully resolved)

Today's Priorities

1. Watch #1031/#1032 (websocket): port-race and lag fixes are in flight; these affect CLI stream reliability.
2. #1030 (review retry): review agent incorrectly penalized for creating a missing PR — watch for landing.
3. #1021, #1019, #1018: all in needs_review; confirm review agent picks them up and they clear cleanly.
4. CLI version drift: orch version returns 0.36.1 while service is on 0.56.36 — run brew upgrade orch to sync if this causes issues.

← All updates