Summary

Strong 24h with 22 commits landed, focused on routing reliability (reroute skipping failed agents, GraphQL error surfacing, GitHub 5xx transient handling) and system hardening (panic fixes, session auto-cleanup). Zero open GitHub issues. The pipeline is healthy with ~82% agent success rate (134/163 runs) and no blocked tasks. Outstanding: the opus/ model assignment bug flagged in yesterday's retro has not yet been resolved.

Recent Activity (Last 24h)

Routing & Dispatch Reliability (4 fixes)

• fix: reroute should skip the previously failed agent — #1493
• fix: clear both agent and model on reroute — #1492
• fix: add error logging to store_reset_failure_counters — #1475
• fix: reset no_code_reroutes on auto-unblock and fix wrong field name — #1475

GraphQL & GitHub API (4 fixes)

• fix: surface GraphQL partial errors in batch_is_pr_merged_by_branch — #1485
• fix: remove invalid escaping in batch GraphQL queries — #1480
• fix: skip transient GitHub 5xx errors in review subscriber — #1484
• fix: surface GraphQL errors and log response body — #1483

System Reliability (5 fixes)

• fix: replace all unsafe .get() with .try_get() in store layer — #1472
• fix: prevent overflow in exponential backoff — #1463
• fix: mark_cleaned() failures now logged — #1477
• fix: avoid blocking Tokio worker in free_models() cache refresh — #1479
• feat: auto-cleanup sessions and worktrees on task close — #1460

Agent Transport & Sessions (3 fixes)

• fix: double-push in TmuxChannel/CaptureService transport — #1464
• feat: one-shot --session-id for chat sessions — #1464
• fix: use --append-system-prompt-file for chat persistent sessions — #1463

Operational Health

Agent Performance (last 24h, 163 total runs)

Overall: ~82% success (134 success / 163 runs). Opencode/opus is the only notable cluster of failures (6 failures, all failed outcome) — tied to the opus/ bug.

Task Activity (last 12h)

• 872 status changes, 229 dispatches, 186 pushes, 162 branch deletes
• 103 review starts, 93 review decisions, 75 PR creates
• 40 errors (low relative to throughput)
• 1 auto-unblock

Orch Doctor Warnings

• #1500: label mismatch — SQLite=in_progress, GitHub labels=none. Non-blocking, likely a sync lag.
• Orphan worktree: /Users/gb/.orch/worktrees/bean/internal-31298-morning-briefing-daily-focus-plan — no task owns it. Created by bean morning briefing job but task appears done/cleaned. Should be pruned.

Log Errors

Stale git worktree metadata (brew error log, benign but noisy):

fatal: not a git repository: /Users/gb/Projects/orch/.git/worktrees/gh-issue-1005-...
fatal: not a git repository: /Users/gb/Projects/orch/.git/worktrees/gh-issue-1130-...
fatal: not a git repository: /Users/gb/Projects/orch/.git/worktrees/gh-issue-1251-...

These are old worktree metadata entries from the user's main orch project. The git worktree prune at startup runs in ~/.orch/worktrees but not in the user's project directory. Low priority — a prune in the project dir would clear it.

Stuck / Blocked Tasks

No blocked tasks. All current tasks are healthy:

Retro Follow-ups (from 2026-03-31 Evening Retro)

Today's Priorities

1. Fix opus/ model assignment bug — 6 failures yesterday, root cause identified in retro. Most impactful fix remaining.
2. Expand .try_get() coverage for SQLite OOB panics — 10 panics in last 24h across sqlx workers. Despite #1472 landing, the panics continue.
3. Prune orphan bean worktree — internal:31298 worktree at ~/.orch/worktrees/bean/ is orphaned. File issue or clean up.
4. Monitor for new SQLite panics — ongoing issue, needs root cause investigation beyond column count.
5. Stale worktree metadata in main project — git worktree prune in ~/Projects/orch/ would clear the brew error log noise.

Open GitHub Issues

0 open issues — The queue has been clear since yesterday's sweep.

← All updates