PD-295 — Décomposition step 6a¶

Stratégie : séquentiel (4 waves)¶

Projet 100% Python (pas de TypeScript). 9 modules répartis en 4 waves séquentielles.

Wave 1 — Foundation (data-format)¶

Task	Module	Agent	Fichiers	Dépendances
T1	`data-format`	agent-developer	`scripts/lib/data_format.py`, `scripts/lib/jsonl_atomic.py`	aucune

Wave 2 — Indexers (B1 + B2)¶

Task	Module	Agent	Fichiers	Dépendances
T2	`veille-indexer` (B1)	agent-developer	`scripts/collect-veille.py`, `scripts/index-veille.py`	T1
T3	`veille-search`	agent-developer	`scripts/search-veille.py`, `.claude/commands/veille-search.md`	T2
T4	`clarifications-store` (B2)	agent-developer	`scripts/collect-clarifications.py`, `scripts/index-clarifications.py`	T1
T5	`clarifications-search`	agent-developer	`scripts/search-clarifications.py`, `.claude/commands/clarifications.md`	T4

Wave 3 — Scoring + Lifecycle (B3 + B4)¶

Task	Module	Agent	Fichiers	Dépendances
T6	`scoring` (B3)	agent-developer	`scripts/compute-reuse-scores.py`	T1
T7	`lifecycle` (B4)	agent-developer	`scripts/learnings-lifecycle.py`	T6, T1

Wave 4 — Injection + Glue (B5)¶

Task	Module	Agent	Fichiers	Dépendances
T8	`injection` (B5)	agent-developer	`scripts/gov-learnings-inject-unified.py`	T3, T5, T7
T9	`gov-step-0-glue` + `reindex-all`	agent-developer	`.claude/commands/gov-step-0.md` (patch), `scripts/reindex-all.py` (patch)	T8

parallelization¶

mode: sequential
reason: "Projet Python scripts, pas de build TS, 1 seul repo, dépendances linéaires entre waves"
waves:
  - [T1]
  - [T2, T3, T4, T5]
  - [T6, T7]
  - [T8, T9]

Contraintes¶

Tous les scripts sont des clones/adaptations de scripts/index-learnings.py, scripts/search-learnings.py, scripts/collect-specs.py
Embeddings via Ollama nomic-embed-text (768 dim) sur IA-Server 192.168.1.82
FAISS index format identique à l'existant (data/learnings-index.faiss)
Pas de TypeScript, pas de build, pas de linter — validation = python3 script.py --help retourne 0