Morning Review — 2026-05-18

2026-05-18

Recent Commits (last 24h)

Hash	Message
`9baf43f8`	docs(posts): add evening retrospective for 2026-05-17 (#3155)
`2481b5cd`	fix(engine): summarise api_retry fragments before persisting task_runs.error (#3154)
`8fa070c2`	fix(cooldown): detect GLM/MiniMax 'Insufficient balance' as credit exhaustion (#3153)
`f4d207cf`	fix(router): proactively filter stale opencode model_map entries (#3152)
`5f303b4e`	docs(posts): morning review for 2026-05-17 (internal:149773) (#3148)

Overall throughput is healthy, but two recurring operational patterns remain visible in logs:

cleanup: timed out listing all tasks for closed-issue reconciliation every sync cycle, followed by using fallback tasks for closed-issue reconciliation.
multi-agent degradation detected repeatedly flagging kimi, minimax, and glm as degraded via generic agent_error cooldown state.

Task activity remains active (last 12h):

status_change=426, dispatch=135, push=111, branch_delete=96, review_start=64, review_decision=55, pr_create=51, error=24.

#3110 (open, blocked): Claude 401 invalid authentication credentials (owner input/logs still required).
internal:149337 (blocked): SSH agent signing failure during push (sign_and_send_pubkey); owner-side SSH agent/key fix still required.

No additional owner-waiting tasks were surfaced in the current open issue list.

Top outcomes indicate generally stable execution with concentrated failures in known pools:

High success volume: claude/sonnet (34), codex/gpt-5.3-codex (23), kimi/opus (16), opencode/github-copilot/gpt-5-mini (11).
Failures/rate limits cluster in degraded pools: kimi/opus failed (4), minimax/opus failed (3), glm/opus rate_limit (3), opencode/github-copilot/gpt-5-mini failed (3).
One stale alias failure still present in history: opencode/github-copilot/gpt-5.3 failed (1) (covered by yesterday’s router filtering fix).

Status of yesterday’s priorities:

Deploy latest releases (v0.71.16–v0.71.18) to activate merged fixes in runtime: still pending confirmation from this task context.
Confirm reconciliation timeout disappearance after deploy: not resolved; warnings are still present in current log output.
Investigate blocked opencode tasks from yesterday: superseded by current generalized degradation/cooldown pattern; monitor remains necessary.
Owner follow-up on #3110 and internal:149337: still pending owner action.

Verify runtime version is fully upgraded to include #3152/#3153/#3154, then re-check whether reconciliation timeout warnings persist.
Triage the root cause of repeated reconciliation list timeout if it continues post-upgrade (query path and timeout budget).
Continue monitoring degraded agent pools (kimi/minimax/glm) and validate cooldown recovery behavior vs repeated agent_error states.
Unblock owner-dependent items: gather concrete auth diagnostics for #3110 and resolve SSH agent/key health for internal:149337.

None.

No new root-cause bug was filed in this pass because the observed operational problems are either:

already tracked (#3110),
owner-environment blockers (internal:149337), or
potentially already addressed by very recent merged fixes pending runtime verification.

Prepared by Orch automation (internal:149847).