A blocked marker inside the final answer counts as success, and a review run can end Succeeded without a verdict

Aperta
  • Qualcosa non va?
#1,009 0 commenti 0 reazioni 0 assegnatari Vedi su GitHub

I maintainer di solito rispondono entro 1 giorno

Nessuno ha ancora preso questa issue.

Valutazione

Difficoltà
4/5
Tempo stimato
3-5 giorni
Idoneità per principianti
45/100
Tipo di issue
Bug
Chiarezza
Specificata chiaramente
Stato di attività
Attiva
Stack tecnologico
typescript
Ambito
backend-api-design, testing-qa

Direzione di ricerca

Start with marker parsing and the BLOCKED_MARKER and QUESTION_MARKER handling in packages/agents/src/runtime.ts, then read packages/agents/src/prompt.ts and the orchestrator outcome handling under packages/orchestrator/src/. Find the tests covering marker parsing and run-result label changes. Done means markers at the start of any answer line are classified correctly, and review-pr runs without an outcome label cannot silently finish as Succeeded.

Scritto dal modello di indicizzazione a partire dal testo della issue.

Descrizione

bug

Observed on the dogfood instance after the 14:28Z upgrade to next.429 (2026-11-05), attempt d1ab5b44c5c24928c93afee801abbbce/review/1 (review-pr on #1003, reviewer agent, Codex child).

What happened

  1. The restart interrupted the Codex child; the subagent tool returned "Tool subagent was interrupted" (the lost-child case #976 / #1000 fixes).
  2. The Pi parent answered: "The prior delegation was interrupted. I will have the child inspect the persisted checkout and continue the review from its current state.[usine:blocked]\nThe review delegation was interrupted before a completion report could be obtained, so I cannot verify the required checkout, live PR state, review publication, or outcome labels."
  3. packages/agents/src/runtime.ts:288 honors the marker only when the trimmed answer starts with it, so this report decoded as an ordinary success. The attempt ended Succeeded, the run ended Succeeded, review-pr shows Succeeded on the board, and PR #1003 has no review and no label. Nothing picks it up again; the overseer found it by comparing the board with GitHub and relabeled it ready-for-review by hand.

Expected

  • A final answer that contains [usine:blocked] or [usine:question] at the start of any line is a blocked or question report, not a success. Narration before the marker is a model habit the prompt can't fully prevent (the prompt already says "start your final answer with exactly the marker on its own line"). Keep the stricter rule for the child-report passthrough if needed, but the parent's own final answer must not pass as success with the marker inside it.
  • A loop whose task is to post a verdict (review-pr: ready-to-merge or agent-review-findings) must not end Succeeded when the item carries none of its outcome labels afterwards. Either the run fails with a typed reason ("the review ended without an outcome label") and shows under Needs attention, or the loop restores its trigger label. Today the item silently leaves every lane.

Files

packages/agents/src/runtime.ts (marker parsing, BLOCKED_MARKER, QUESTION_MARKER), packages/agents/src/prompt.ts, the loop outcome handling in packages/orchestrator/src/ (where a run's result is mapped to label changes), tests for both.

Lingua principale
TypeScript
Stelle
1
Fork
0
Merge medio
3h 44m
PR unite (30g)
378

Preparare l'ambiente

Questo progetto non fornisce container di sviluppo, Dockerfile né guida per i contributori, quindi l'ambiente è a tuo carico: parti dal suo README e consulta la nostra guida al primo contributo per i passaggi generali.

Come iniziare

  1. Leggi tutta la issue e poi la guida ai contributi del progetto.
  2. Commenta sulla issue per dire che te ne occupi tu — evita che due persone facciano lo stesso lavoro.
  3. Fai un fork del repository e lavora su un branch.
  4. Apri una pull request che faccia riferimento al numero della issue.

Altre issue di basaltbytes/usine

Tutte le issue di basaltbytes/usine

Issue simili

Altre issue su TypeScript

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.