Files
Kill_LIFE/docs/AGENTIC_LANDSCAPE.md
Clément SAILLANT 5efb380da5 feat: automate yiacad PR review lane (#18)
* feat: automate yiacad pr review lane

* fix: stabilize yiacad review and spec gates

* fix: scope kicad exports to repo hardware

* fix: ignore kicad block library sources in exports

* feat: Add infra VPS monitoring runbook and healthcheck scripts

- Created INFRA_VPS_RUNBOOK_2026.md detailing operational procedures for monitoring VPS services.
- Added infra_vps_healthcheck.sh script for automated health checks on DNS, TLS, TCP, and HTTP for VPS services.
- Introduced infra_vps_security_audit.sh for non-intrusive security checks on external VPS services.
- Established JSON schema for infra VPS inventory in infra_vps.schema.json.
- Developed integration for runtime status reporting in the Next.js API route.
- Implemented Playwright tests for smoke testing the application and ensuring core functionalities.
- Updated Makefile for development dependencies and testing commands.
- Created various test files for unit and end-to-end testing across different components.

* feat(agentics): update mesh agents, gates, prompts, and workflows
2026-03-29 17:14:16 +02:00

5.9 KiB
Raw Permalink Blame History

Agentic Landscape (Kill_LIFE)

Posture de départ

  • Spec-driven backbone: specs/* et specs/constraints.yaml
  • Standards and gates: compliance/*, bmad/*
  • Runtime local: ZeroClaw + MCP + crazy_life comme éditeur
  • Overlays: LangGraph, AutoGen, n8n en option contrôlée
  • Exécution locale de préférence, GitHub pour la traçabilité distante
flowchart TD
  Inputs[Issue / PR / demande humaine] --> Labels[Labels ai:* + Scope]
  Labels --> Router{Routeur dexécution}
  Router --> SpecPath[specs/ (00->04)]
  Router --> GitHubPath[workflow local / GitHub dispatch]

  SpecPath --> Validation{specs + compliance}
  Validation --> Tools[tools/*]
  Tools --> RepoTools[repo_state + compliance + auto_check_ci_cd]
  Tools --> MCP[tools/hw/* + MCP launchers]
  Tools --> TUI[tools/cockpit/refonte_tui.sh]
  Tools --> CADFusion[tools/cad/yiacad_fusion_lot.sh]

  CADFusion --> CadEvidence[Evidence CAD (cad-fusion)]
  MCP --> Evidence[Evidence packs locaux]
  RepoTools --> Evidence
  CadEvidence --> Evidence
  TUI --> Logs[artifacts/refonte_tui/*.log]
  Logs --> Evidence
  Evidence --> Audit[docs/evidence/*]
  GitHubPath --> CI[.github/workflows/*]
  CI --> Audit
  Audit --> PM[PM + Plan]
  PM --> Handoff[Handoffs / Plans / Labels]
  Handoff --> Agents[agents/ + .github/agents/ + prompts]

Modèle de fonctionnement

  • agents/ et .github/agents/ définissent les compétences minimales.
  • tools/cockpit/refonte_tui.sh est la surface de pilotage texte (status, logs, lot-chain).
  • ZeroClaw reste le runtime local dorchestration, sans remplacer la source de vérité.
  • Les surfaces MCP sont locales par défaut, huggingface en option distante officielle.
  • Les preuves et traces sont obligatoires pour les lots de refonte.

Décisions de refonte actives

  • pas de dépendance forte externe pour les lanes essentielles,
  • tout overlay IA passe par une validation de sécurité (labels, scope, denylist),
  • priorité à la reproductibilité documentaire (specs, plans, diagrams, evidence).

Delta mesh 2026-03-20

Le paysage agentique du projet opere desormais comme un overlay de gouvernance maillée sur trois repos, avec neuf agents principaux.

Agent Repo owner prefere Responsabilite
PM-Mesh Kill_LIFE backlog tri-repo, priorisation, ordre de propagation
Arch-Mesh Kill_LIFE contrats inter-repos, ADR, compatibilite
Embedded-CAD Kill_LIFE firmware, hardware, evidence embarquee
Runtime-Companion mascarade runtime IA, providers, adaptateurs MCP
CAD-Fusion Kill_LIFE pilote YiACAD, health CAD, preuve et rollback
Web-Cockpit crazy_life UI cockpit, workflow editor, supervision
QA-Compliance transversal matrices, smokes, preuves
Docs-Research transversal README, specs, Mermaid, benchmark OSS
SyncOps Kill_LIFE SSH, repo-state, TUI, logs, sync continue

Sous-agents standardises:

  • Lot-Planner, Risk-Triager
  • Schema-Guard, Compat-Auditor
  • Firmware-Lane, CAD-Lane, CAD-Fusion
  • Provider-Bridge, Runtime-Smoke
  • UI-Flow, Schema-Consumer
  • Spec-Mirror, Evidence-Pack
  • Mermaid-Map, OSS-Benchmark
  • SSH-Health, Log-Ops

Skills opératoires actives:

  • bash-cli-tui: SyncOps, CAD-Fusion, Doc-Runbook
  • playwright: Web-Cockpit quand une validation UI réelle devient nécessaire
  • OSS-Benchmark: Docs-Research, CAD-Fusion
  • Evidence-Pack: QA-Compliance, Docs-Research

Routing prioritaire des lots ouverts:

  • PM-Mesh garde T-RE-204 comme lot exécutable bloquant.
  • CAD-Fusion tient la lane parallèle T-RE-209 / T-RE-210.
  • SyncOps publie la synthèse hebdomadaire et la checklist de sortie (T-RE-301 / T-RE-304).

Regle de coexistence: nous ne sommes pas seuls dans le codebase; toute intervention doit rester dans un write_set declare et observable.

Matrice detaillee publiee: docs/AGENT_SPEC_MODULE_MATRIX_2026-03-20.md

Delta d'execution machine-readable:

  • Schema-Guard publie des schemas JSON versionnes pour agent_handoff, repo_snapshot, workflow_handshake.
  • Runtime-Smoke et Schema-Consumer consomment ces contrats depuis la lane soeur Kill_LIFE-main au moment des preflights compagnons.
  • La validation CI visible du handshake passe desormais par .github/workflows/mesh_contracts.yml.

Delta agentics 2026-03-29

Mise a jour de la lane PR review automation (branche feat/yiacad-pr-review-automation, PR #18).

Nouveautes

  • Gate S2 (bmad/gates/gate_s2.md) -- couvre la lane Web & PR review : Playwright, BullMQ worker, evidence intelligence_program/, validation summary_short.schema.json.
  • Subagent PR-Review-Orchestrator -- ajoute a Web-CAD-Platform, write set : web/project/.ci/.
  • Schema-Guard -- reference specs/contracts/yiacad_uiux_output.schema.json comme contrat evidence de la lane review.
  • QA-Compliance -- gate S2 ajoutee aux gates de validation release.
  • bmad/templates/handoff.md -- enrichi : sections Gate passed + Consumer contracts.
  • bmad/rituals/kickoff.md -- identification dynamique des gates (S0/S1/S2/...) selon le domaine du lot.

Table agents canoniques (2026-03-29)

Agent owner_agent Handoff contract Gates
PM-Mesh Kill_LIFE agent_handoff.schema.json S0
Arch-Mesh Kill_LIFE agent_handoff.schema.json S0
Schema-Guard Kill_LIFE workflow_handshake.schema.json S0
QA-Compliance transversal operator_lane_evidence.schema.json S0, S1, S2
KillLife-Bridge Kill_LIFE agent_handoff.schema.json S0
Runtime-Companion Kill_LIFE agent_handoff.schema.json S0
SyncOps Kill_LIFE operator_lane_evidence.schema.json S0
Web-CAD-Platform Kill_LIFE summary_short.schema.json S0, S2
Docs-Research transversal agent_handoff.schema.json S0
Embedded-CAD Kill_LIFE agent_handoff.schema.json S0, S1
Firmware Kill_LIFE operator_lane_evidence.schema.json S1
UX-Lead Kill_LIFE agent_handoff.schema.json S0