feat: sync moteur vers OSS — mise à jour majeure (2026-08-11)

[Axellum]

Résumé

Synchronisation du miroir public depuis le dépôt privé \moteur_agents\ (procédure \prepare_oss_export.ps1\ + anonymisation).

Contenu

• core/ : +14 modules (env_bootstrap, provider_plugin, agent_trace, log_redaction, model_inventory, dag/node_tier_policy, ha_token, fenetres_execution, builtin_providers, fallback_trace, ...) + mise à jour de ~70 modules (router, llm_gateway, vocal_, mcp_, tab5_pusher, watchdog...)
• services/ : +approval_resume_service, +coding_front ; maj execute_service, ha_state_query, ha_vocal_fallback, pipeline_service
• tools/ : +testing, +dag_map_reduce, +registry_setup, +code_front, +db_optimize, +discover_provider_models ; maj terminal, tool_registry, sanitizer
• tests/ : +88 tests (routage session_id, budget, providers async, vocal STT/TTS, traçabilité T296, dag node tiers...)
• scripts/ : +gen_ha_local_fallback (répare le test déjà présent sur main)
• docs/ARCHITECTURE.md : à jour
• seed_models_db.py, gui_server.py, workspace_mcp.py, requirements, pyproject : à jour

Sécurité (repo public)

• Scan \oss_secrets_scan.py\ : ✅ aucun secret / IP LAN réelle
• Anonymisation appliquée : IPs LAN → \ etc., entités HA réelles → entités neutres (light.salon → light.living_room...) dans le code, \ha_commands.json.example\ et les YAML deploy régénérés
• Exclus du miroir : ihm-v2/, .env, deploy/.env, memory/episodes/, tools/llama.cpp, outils internes (datasets, finetune, vertex, billing)
• Aucune suppression de fichiers existants

Validation

• \pytest tests/\ : ✅ vert en local (exit 0) — suite complète
• \oss_secrets_scan.py\ : ✅
• Seed models DB : ✅

---

Note

High Risk
Very large sync touching planner/executor/reviewer/HITL, tool execution permissions, and auth paths—high blast radius despite many targeted production bugfixes.

Overview
Large public mirror sync that hardens the DAG / agent loop and expands operator-facing APIs, plus new GitHub community files (bug/feature issue templates, Contributor Covenant CoC, README CI badges).

Orchestration & agents: BaseAgent now wraps each subclass invoke() with a ContextVar agent trace (#T296) so token_usage.agent_name is populated. Executor gains per-agent tool allowlists, passes agent_name into the registry, optional project conventions on LLM calls, post-write_file test runs, and T274 logic so verify/read tasks are not falsely failed for “no tool called”. Planner adds mandatory criteres_acceptation, parallel-stage guidance, tier normalization, configurable planner timeout (#T277), and externalized prompts. Reviewer fixes git-based auto-approve (#T209/#T276) and forces LLM review when git is unreliable. ToolMaker drops local subprocess sandbox for GitHub Actions remote validation (#T207) and persists generated minimal tests (#T190).

API & security: HITL pending/approve/reject reads a durable store and resumes or cancels DAGs in background (#T267). Routing passes session_id into analyze_request. Custom agents expose allowed_tools in CRUD. Models registry adds inventory/offered-by-API, bulk enable/disable, ping, routing-tier updates; batch routes removed from external APIs. WebSocket auth uses shared require_websocket_auth; HA uses centralized get_ha_token. Anthropic billing reconciliation endpoint added.

^{Reviewed by Cursor Bugbot for commit 64d7ec2. Configure here.}

[cursor]

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: serverGenReqId_b492df66-d989-491b-9ec2-739dd4759bc5)

[Axellum]

Re-déclenchement CI