Claude Code Plugins

Community-maintained marketplace

Feedback

output-conventions

@nera0875/agi
0
0

Output structure - fichiers artifacts organisés par scope

Install Skill

1Download skill
2Enable skills in Claude

Open claude.ai/settings/capabilities and find the "Skills" section

3Upload to Claude

Click "Upload skill" and select the downloaded ZIP file

Note: Please verify skill by going through its instructions before using it.

SKILL.md

name output-conventions
description Output structure - fichiers artifacts organisés par scope

Output Convention - Fichiers Structurés

Principe : Outputs organisés par scope (projet, plugin, memory) avec validation stricte.

Structure Obligatoire

Projet/Plugin (.plan/) :

.plan/
├── requirements.md      # Besoins listés
├── workflow.yaml        # Phases séquentielles
├── architecture.md      # Design decisions
├── tasks.md            # Tracking checkboxes + timestamps
├── state.json          # État dynamique (current_phase, agents_history)
└── roadmap.yaml        # Milestones (optionnel)

Memory (Persistant) :

.claude/
├── context.json        # Consolidated memory (writor + daemon update)
└── data/brain/         # Brain-specific data (TickTick sync, energy logs)

Plugins Structure :

.claude-plugin/
├── plugin.json         # Manifest (name, version, agents, skills, commands)
├── agents/*.md         # Agents (≤30 lignes chacun)
├── skills/*/*.md       # Skills (≤50 lignes implémentation)
└── commands/*.yaml     # Commands metadata

Règles Sorties

CRÉER TOUJOURS :

  • .plan/state.json (source vérité workflow)
  • .plan/tasks.md (tracking human-readable)
  • Architecture si >3 agents ou décisions clés

JAMAIS CRÉER :

  • README.md (documentation = context.json)
  • REPORT.md, SUMMARY.md (artifacts parasites)
  • .backup, .tmp, .old (Git = historique)
  • Fichiers temporaires hors scope défini

Validation Output

Avant finaliser :

  1. Grep context.json (vérifier pas doublon)
  2. Glob scope (vérifier pas fichier parasite)
  3. Validate JSON/YAML (strict-validation skill)
  4. Update state.json (final step)

Format Agents History (state.json)

{
  "agents_history": [
    {"agent": "writor", "action": "LOAD context", "timestamp": "2025-10-27T14:30:00Z"},
    {"agent": "executor", "action": "Created skill X", "timestamp": "2025-10-27T14:35:00Z"}
  ]
}

Orchestrator responsabilité : Update state.json après CHAQUE agent complété.

Intégration Workflows

Phase Architect → Outputs : .plan/architecture.md + state.json Phase Executor → Outputs : Agents/skills créés + state.json update Phase Validation → Outputs : Zero artifacts (validation inline)