Dom/rpa_vision_v3
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

docs(coordination): report c gamma dashboard promotion

Browse Source
This commit is contained in:
Dom
2026-05-29 21:49:36 +02:00
parent 34527b5cc5
commit f2a9e40502
3 changed files with 182 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

82
docs/coordination/syntheses/2026-05-29_2148_codex_RESULTAT-C-gamma-dashboard-promotion-dryrun.md Normal file
View File

@@ -0,0 +1,82 @@
# Codex - RESULTAT C-gamma dashboard promotion dry-run no-CLI
Date: 2026-05-29 21:48
Branche: `backup/post-demo-2026-05-19`
## Sources
Retours lus:
- `docs/coordination/inbox_codex/2026-05-29_2140_claude-to-codex_ACK-mission-Cgamma-dashboard-no-cli.md`
- `docs/coordination/inbox_codex/2026-05-29_qwen-to-codex-claude_REVUE-CONTRAT-C-GAMMA.md`
Decision Dom appliquee:
- pas de CLI operateur en demo/POC;
- promotion/write-back uniquement via dashboard avec bouton;
- dry-run + preview + confirmation;
- pas de commit git automatique depuis backend.
## Implementation
Commit:
- `34527b5cc feat(lea): add dashboard competence promotion dry run`
Ajoute:
- `core/competences/promotions.py`;
- endpoint VWB/backend `POST /api/v1/lea/competences/<competence_id>/promote`;
- endpoint dashboard 5001 meme route;
- stats `/api/knowledge-base/stats` enrichies avec `competences`;
- section dashboard `/knowledge-base` "Competences apprises par supervision";
- boutons de promotion avec dry-run, modal diff YAML, evidence verdicts, confirmation explicite;
- audit `data/competences/promotions.jsonl` seulement au confirm;
- backup YAML timestamp + atomic write;
- aucun commit git automatique.
## Contrat
Dry-run:
- `dry_run=true`;
- ne modifie aucun YAML;
- retourne `dry_run_token`, `yaml_diff`, `eligible`, `blocking_reasons`.
Confirm:
- `dry_run=false`;
- exige le `dry_run_token` exact;
- refuse si non eligible;
- backup YAML;
- write atomique;
- validation YAML post-write;
- append `promotions.jsonl`.
Regles:
- `observed -> candidate`: 1 verdict `valid` avec `workflow_id` et `step_results[]`;
- `candidate -> stable`: 3 verdicts valid, 3 contextes distincts, 0 invalid non explique;
- pas de downgrade automatique;
- regression suspectee exposee au dashboard.
## Validations
Tests:
- `.venv/bin/python3 -m pytest tests/unit/test_competence_promotions.py tests/unit/test_lea_competence_verdict_api.py tests/unit/test_dashboard_routes.py tests/unit/test_dashboard_auth_p0a.py tests/unit/test_competence_verdicts.py tests/unit/test_competence_catalog_loader.py tests/unit/test_competence_to_vwb_preview.py tests/unit/test_supervised_popup_guard.py tests/unit/test_vwb_supervised_pause_runtime.py tests/unit/test_vwb_wait_for_state.py -q`
- resultat: 82 tests OK.
Controles:
- `py_compile` OK;
- `git diff --check` OK;
- parsing JS dashboard OK via Node.
Smoke live:
- serveurs actifs:
- dashboard `http://localhost:5001/knowledge-base`;
- backend VWB `http://localhost:5002`;
- frontend VWB `http://localhost:3002`;
- `GET /api/knowledge-base/stats` -> 200 avec 9 competences (`candidate`: 6, `observed`: 3);
- dry-run dashboard promote -> 200, `eligible=false`, `yaml_write=false`;
- dry-run VWB backend promote -> 200, `eligible=false`, `yaml_write=false`;
- aucun YAML competence modifie;
- aucun `data/competences/promotions.jsonl` cree par les dry-runs.
## Limites connues
- Les boutons restent naturellement desactives tant qu'il n'existe pas assez de verdicts `valid` avec evidence.
- L'agent de menage worktree n'est pas implemente dans C-gamma; chantier separe.
- Le diff YAML est produit depuis une serialisation PyYAML complete, donc il peut etre plus large qu'un patch texte minimal; le write reste valide, audite et reversible.