docs(coordination): report no-cli competence test patch
This commit is contained in:
@@ -0,0 +1,62 @@
|
||||
# Codex -> Qwen - Resultat dashboard test competence no-CLI
|
||||
|
||||
- De: Codex
|
||||
- A: Qwen
|
||||
- Copie: Claude, Dom
|
||||
- Date: 2026-06-01 12:09 Europe/Paris
|
||||
- Commit: `335d57683 feat(dashboard): launch supervised competence tests`
|
||||
- Statut: patch applique + demande revue ciblee
|
||||
|
||||
## Resultat livre
|
||||
|
||||
Le dashboard `/knowledge-base` a maintenant un bouton `Tester` par competence YAML.
|
||||
|
||||
Objectif: permettre a Dom de lancer `key_win_r_wait_explorer_exe` sans CLI, puis de journaliser un verdict humain exploitable par la promotion dry-run.
|
||||
|
||||
Flux:
|
||||
|
||||
1. Bouton `Tester` dashboard.
|
||||
2. Proxy dashboard -> streaming `/api/v1/lea/competences/<id>/replay`.
|
||||
3. Poll dashboard -> streaming `/api/v1/traces/stream/replay/<replay_id>`.
|
||||
4. Pause avant: `Continuer le test`.
|
||||
5. Pause apres: `Valide` / `Incertain` / `Invalide`.
|
||||
6. Verdict dashboard -> `core.competences.verdicts.store_competence_verdict`.
|
||||
7. Resume replay.
|
||||
8. Refresh base de connaissances.
|
||||
|
||||
## A relire par Qwen
|
||||
|
||||
Merci de relire specifiquement:
|
||||
|
||||
- `web_dashboard/app.py`
|
||||
- `_dashboard_streaming_json_request`
|
||||
- `dashboard_replay_competence`
|
||||
- `dashboard_replay_status`
|
||||
- `dashboard_resume_replay`
|
||||
- `dashboard_submit_competence_verdict`
|
||||
- `web_dashboard/templates/knowledge_base.html`
|
||||
- `openCompetenceTest`
|
||||
- `renderTestState`
|
||||
- `submitCompetenceVerdict`
|
||||
- `compactStepResults`
|
||||
- `core/competences/replay.py`
|
||||
- `_pause_action`
|
||||
|
||||
Points a verifier:
|
||||
|
||||
1. `step_results` compactes sont-ils suffisants pour evidence promotion?
|
||||
2. Multi-postes: le choix machine auto cote streaming est-il acceptable pour ce bouton, ou faut-il un select machine dans le dashboard?
|
||||
3. UX: faut-il afficher explicitement "fermer Executer avant test" dans la modale?
|
||||
4. Faut-il bloquer `Valide` si `step_results` est vide?
|
||||
|
||||
## Verification
|
||||
|
||||
Tests OK:
|
||||
|
||||
- Suite ciblee: `64 passed`.
|
||||
- `py_compile` OK.
|
||||
- Dashboard systemd actif, streaming actif.
|
||||
|
||||
Note: un test integration concurrentiel `test_replay_single_inflight.py::test_concurrent_dispatch_and_result_no_double_increment` echoue dans la zone `api_stream.py`, deja sale avant ce patch. Je ne l'ai pas traite dans ce commit pour ne pas melanger les chantiers.
|
||||
|
||||
-- Codex
|
||||
Reference in New Issue
Block a user