Hacktoberfest 2026: le issue che i maintainer hanno segnato per ottobre, aperte e adatte ai principianti. Sfoglia le issue Hacktoberfest

[Bug]: Human-criterion classification only matches English keywords, so in a non-English project every review goes to an AI reviewer and the Awaiting review column (#92) stays empty

Aperta
  • Qualcosa non va?
#102 4 commenti 0 reazioni 0 assegnatari Vedi su GitHub

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
Abbastanza chiara
Stato di attività
Attiva
Stack tecnologico
github, python, shell
Ambito
ai-infra-agents, cli

Direzione di ricerca

Start with _ACT_IN_THE_WORLD_RE and the criterion classification in lib/delegation.py; trace how [RUBBER-STAMP] and [REVIEWER] are handled, then check review_readiness in web/shared.py and verdict handling in lib/verdict_ledger.py. Reproduce with the French fw task classify-ac examples. Done when an explicit, language-independent operator-only classification works and those criteria do not require an AI verdict for the review queue.

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

Descrizione

area:governance bug severity:medium status:in-progress triaged
Bug Description

lib/delegation.py decides who may close a Human criterion. The operator-only class act-in-the-world is detected by _ACT_IN_THE_WORLD_RE (line 188), a list of English words: publish, deploy, production, git push, customer, payment... A criterion written in another language matches nothing, falls to unclassified, and unclassified is reviewer-judged (T-3557).

Since T-4053, a task enters the "Awaiting review" column (#92, thank you for it) only when every reviewer-judged criterion has a green verdict from an independent reviewer (web/shared.py, review_readiness; lib/verdict_ledger.py). That reviewer is another AI agent dispatched by the framework.

In a project written in French, the operator's own review steps ("the operator reviews deliverable A and presents it to the client's founder") are therefore never operator-only. The column shows a task only by accident: here, one task out of seven appears, because its text happens to contain an English phrase the regex knows.

The criterion author can declare a criterion mechanical ([RUBBER-STAMP], [REVIEWER], lines 246-248 and 583-585), but there is no way to declare it operator-only.

Steps to Reproduce
  1. fw task classify-ac --text '- [ ] [REVIEW] l’opérateur relit la trame du chantier A, la présente à la fondatrice du client au point du 2026-10-21 et recueille sa validation'
    -> REVIEWER-JUDGES (unclassified) — no deterministic signal — an independent reviewer judges it
  2. fw task classify-ac --text '- [ ] [REVIEW] l’opérateur remet le livrable A au client (dépôt GitHub du client) après relecture'
    -> same result, although handing a deliverable to a client is an act outside the repository.
  3. fw review-queue: one task "ready", six "waiting on the agent" with "independent review not green on criterion N". All seven have their Agent criteria ticked and wait for the operator.
Expected Behavior

A way to classify a criterion that does not depend on English: for example an explicit [OPERATOR] prefix symmetrical to [RUBBER-STAMP] / [REVIEWER], or the project-level declaration discussed in T-3968. A task whose open Human criteria are operator-only then enters "Awaiting review" without an AI verdict.

Actual Behavior

Every Human criterion of a non-English project is sent to an AI reviewer. The operator who does not want deliverables judged by another AI (confidentiality, client sign-off) cannot use the "Awaiting review" column at all.

fw doctor output
AI Agent

Claude Code

Operating System

Linux

Additional Context

1.8.11 (e42aa41bf), vendored consumer, Linux (Ubuntu 20.04), non-root account, Claude Code. We decided not to rephrase our criteria with English keywords just to pass the regex: that would mislead the classifier rather than describe the work.

Lingua principale
Shell
Stelle
16
Fork
6
Merge medio
6h 24m
PR unite (30g)
5

Preparare l'ambiente

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 DimitriGeelen/agentic-engineering-framework

Tutte le issue di DimitriGeelen/agentic-engineering-framework

Issue simili

Altre issue su Shell/Bash

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.