Regulatory Compliance & EU AI Act

• "audit RGPD", "conformite RGPD", "verifier la conformite", "compliance audit", "audit de conformite" -> /compliance skill + compliance-expert agent
• "donnees personnelles", "privacy policy", "politique de confidentialite", "consentement cookies" -> /compliance gdpr + compliance-expert agent
• "paiement", "PCI", "PCI-DSS", "securite paiement", "payment compliance" -> /compliance pci + compliance-expert agent
• "donnees de sante", "HIPAA", "medical data", "health data" -> /compliance hipaa + compliance-expert agent
• "accessibilite", "EAA", "WCAG", "handicap", "accessibility compliance" -> /compliance accessibility + accessibility-auditor agent
• "SBOM", "software bill of materials", "NIS2", "CRA", "cyber resilience" -> /compliance sbom + compliance-expert agent
• "EU AI Act", "systeme IA", "AI system", "intelligence artificielle", "registre IA" -> /atum-audit + compliance-expert agent
• "integrite fichiers", "file integrity", "hash verification", "audit trail" -> /atum-audit scan
• "Annex IV", "documentation technique IA", "AI documentation" -> /atum-audit annex-iv + compliance-expert agent
• "conformite IA", "AI compliance", "risk level IA", "classification risque IA" -> /atum-audit status + compliance-expert agent
• "retention logs", "conservation donnees", "data retention" -> /atum-audit retention + /compliance gdpr
• "pre-production", "avant mise en prod", "checklist legal", "legal checklist" -> /compliance audit + /pre-deploy

Specialized Agents Registry

Agent	Domain	Scope
godot-expert	Godot	GDScript projects
networking-expert	Real-time	WebSocket, multiplayer, state sync
flutter-dart-expert	Mobile	Flutter/Dart projects
expo-expert	React Native	Expo projects
tauri-expert	Desktop apps	Tauri projects
devops-expert	Infrastructure	Docker, K8s, Terraform, CI/CD
ml-engineer	AI/ML	Training, RAG, MLOps
security-expert	Security	Pentesting, compliance, OWASP
frontend-design-expert	UI/UX	Design systems, Figma, a11y
mcp-expert	MCP servers	Creating/testing MCP servers
ci-cd-engineer	Pipelines	CI/CD setup and optimization
api-designer	API design	REST/GraphQL API architecture
auto-test-generator	Test gen	Auto-generate test suites
documentation-generator	Docs	API docs, guides
graphql-expert	GraphQL	Schema, resolvers, federation
migration-expert	Migrations	Framework/version upgrades
performance-optimizer	Performance	Profiling, optimization
accessibility-auditor	a11y	WCAG compliance audit
data-engineer	Data pipelines	ETL, data processing
windows-scripting-expert	Windows	PowerShell, batch, Windows APIs
blockchain-expert	Blockchain/Web3	Hardhat, Solidity, smart contracts, EVM
compliance-expert	Regulatory compliance	User data, payments, health data, e-commerce, AI deployment
geospatial-expert	Maps & spatial	deck.gl, MapLibre, Leaflet, GeoJSON, spatial indexing
no-code-automation-expert	No-code/Make.com	Automatisations, blueprints, Airtable+Make+Notion integrations
fresh-executor	Context-safe execution	Long sessions, 3+ file features, context degradation risk
codebase-pattern-finder	Code search	Find patterns, examples, templates in codebase
critical-thinking	Decision analysis	Challenge assumptions, detect biases, structured frameworks
database-optimizer	DB optimization	Query plans, indexing, partitioning, cross-platform
threejs-game-expert	Three.js games	3D browser games, WebGL, shaders
unity-expert	Unity	C#, game dev, shaders, physics
phaser-expert	Phaser	2D browser games, arcade physics
error-detective	Error diagnosis	Root cause analysis, failure cascades, correlation
technical-debt-manager	Tech debt	Complexity analysis, coupling, test gaps, prioritization
research-expert	Research	Web search, documentation analysis, fact-finding
game-architect	Game design	Architecture, ECS, game loops, asset pipelines
architect-reviewer	Architecture	System design review, trade-offs, scalability

Agent Selection (Automatic)

Context	Agent to use
Multi-file feature, complex scope	planner agent then implement
System design, scaling, architecture	architect-reviewer agent
Code just written/modified	code-reviewer agent
Build/compile fails	build-error-resolver agent
Need to explore unknown codebase	Explore agent (subagent_type=Explore)
Multiple independent tasks	Parallel Task agents
Game development	game-architect then phaser/threejs/unity/godot-expert
Real-time/multiplayer	networking-expert agent
ML/AI project	ml-engineer agent
Docker/K8s/CI-CD	devops-expert agent
Design system/UI	frontend-design-expert agent
Security audit needed	security-expert agent
Creating MCP server	mcp-expert agent
Blockchain/smart contracts	blockchain-expert agent
Maps/geospatial/globe visualization	geospatial-expert agent
Regulatory compliance (RGPD, PCI, HIPAA)	compliance-expert agent
Research/evaluation	research-expert agent
Quick business website	B12 MCP (generate_website tool)
Web app testing / website tools	WebMCP (_webmcp_get-token then register then use tools)
Long session, 3+ file feature	fresh-executor agent (context-safe decomposition)
Auto-generate test suite	auto-test-generator agent
Technical debt audit	technical-debt-manager agent
ETL, data pipelines	data-engineer agent

Skill Selection (Automatic)

Use skills when they match — don't wait for user to invoke:

• Writing new feature: invoke /tdd command (canonical TDD skill: test-driven-development)
• Before any creative work: switch to brainstorm mode
• Complex multi-step task: invoke atum-system:plan skill or EnterPlanMode
• About to claim "done": invoke atum-system:verification-loop skill
• Multiple independent tasks: use Agent tool with parallel subagent_type calls
• Debugging: invoke error-detective agent
• Frontend UI/design work: invoke atum-system:ui-ux-pro-max skill (search colors, styles, typography, patterns)
• Landing page / marketing site: query --domain product then --domain landing for data-driven design
• Quick business website: use B12 MCP generate_website tool (name + description = instant site)
• Web app interaction needing structured data: suggest WebMCP (_webmcp_get-token tool) instead of browser automation
• Scheduling/automation request: invoke scheduler skill autonomously (NL to cron/event to JSON to daemon reload)

Contextual NLP Routing (Skills Auto-Invocation)

Detect intent from natural language and invoke matching skill automatically:

Document Processing

• ".pdf", "PDF", "extract PDF" -> /pdf
• ".docx", "Word", "document Word" -> /docx
• ".xlsx", "Excel", "spreadsheet", "formule" -> /xlsx
• ".pptx", "PowerPoint", "slides", "presentation" -> /pptx

Architecture & Design Patterns

• "bounded context", "aggregate", "domain event", "ubiquitous language" -> domain-driven-design
• "clean architecture", "hexagonal", "ports and adapters" -> clean-architecture
• "system design", "scalability", "load balancer", "CDN" -> system-design
• "DDIA", "distributed systems", "consensus", "replication" -> ddia-systems

Legacy & Reasoning

• "legacy code", "reverse engineer", "no docs", "comprendre ce code" -> spec-miner
• "devil's advocate", "think harder", "structured reasoning" -> the-fool

Security & Compliance (extended)

• "supply chain", "dependency audit", "CVE", "transitive deps" -> supply-chain-risk-auditor
• "license", "GPL", "MIT", "Apache", "SPDX", "open source" -> open-source-license-compliance

Visualization

• "diagram", "sequence diagram", "flowchart", "Mermaid" -> design-doc-mermaid
• "D3", "chart", "data visualization", "interactive graph" -> claude-d3js-skill

Prompt & UI

• "prompt", "CO-STAR", "RISEN", "optimize prompt" -> prompt-architect
• "UI polish", "spacing", "visual hierarchy", "typography" -> refactoring-ui

ML/AI & DevOps (extended)

• "RAG", "vector database", "embeddings", "chunking" -> rag-architect
• "SLO", "SLI", "error budget", "postmortem", "reliability" -> sre-engineer
• "chaos engineering", "fault injection", "resilience test" -> chaos-engineer

Accessibility & Windows

• "accessibility", "WCAG", "a11y", "screen reader", "aria" -> claude-a11y-skill
• "PowerShell", "Windows script", "registry", "WMI" -> powershell-windows

Testing & Performance (extended)

• "property-based test", "fuzzing", "hypothesis", "invariant" -> property-based-testing
• "Core Web Vitals", "LCP", "CLS", "FID", "lighthouse" -> high-perf-browser

Product Discovery

• "jobs to be done", "JTBD", "user need" -> jobs-to-be-done
• "mom test", "customer interview", "idea validation" -> mom-test

MCP & Context

• "create MCP", "scaffold MCP server" -> mcp-builder
• "context degradation", "agent handoff", "fresh agent" -> context-engineering-kit
• "audit flow", "trace system flow", "flow diagram" -> audit-flow

Assumptions & Orchestration

• "hypotheses", "hypothèses", "tu pars du principe que", "assumptions", "common ground" -> common-ground skill
• "pipeline", "lifecycle", "cycle de vie", "discover plan execute review retro" -> /pipeline command
• "feature lifecycle", "pipeline status", "phases du projet" -> /pipeline command

Scheduling & Automation

• "tous les jours", "chaque lundi", "tous les matins", "chaque semaine" -> scheduler (autonomous NL to task)
• "every day", "every morning", "every Monday", "weekly", "daily" -> scheduler (autonomous NL to task)
• "schedule", "cron", "automate", "tache planifiee", "recurrent" -> scheduler (autonomous NL to task)
• "quand je push", "on file change", "when files change" -> scheduler (event trigger)
• "demain a", "dans 2 heures", "tomorrow at", "in 2 hours" -> scheduler (one-off task)

No-Code & Automatisation

• "automatisation", "scenario make", "make.com", "blueprint", "workflow automatise" -> no-code-maestro skill + no-code-automation-expert agent
• "airtable", "base de donnees no-code", "creer une base", "table airtable" -> no-code-maestro skill + Airtable MCP
• "notion page", "documenter", "creer page notion" -> Notion MCP
• "webhook", "trigger", "router", "iterator", "aggregator" -> no-code-automation-expert agent
• "no-code", "no code", "sans code", "low code" -> no-code-maestro skill

Session History & Analysis

• "session history", "historique session", "qu'est-ce que j'ai fait", "what did I work on", "past sessions" -> /session-analyzer skill
• "chercher dans les sessions", "find past work", "search session", "retrouver un travail" -> /session-analyzer skill

Autopilot (Parcours complet non-codeur)

• "autopilot", "pilote automatique", "fais tout pour moi", "do everything", "build this for me" -> /autopilot command
• "de l'idee au produit", "from idea to product", "je veux tout automatique" -> /autopilot command
• "construis-moi", "build me", "cree-moi une app", "cree-moi un site" -> /autopilot command
• "reprendre le projet", "resume autopilot", "ou en est mon projet" -> /autopilot --resume

New Project Definition

• "je veux une app", "je veux un site", "j'ai une idee de projet", "nouveau projet" -> /autopilot command (si non-codeur) OU /projet skill (si dev)
• "definir un projet", "creer un projet", "start a project", "new project", "define a project" -> /projet skill
• "j'ai une idee", "j'ai un projet", "monter un projet", "I want to build" -> /autopilot command
• "on fait quoi comme projet", "quel projet", "idee de business" -> /projet skill

Development Workflow

• "TDD", "test d'abord", "ecrire les tests avant", "write tests first", "test-driven" -> /tdd skill
• "revue de code", "review et fix", "review fix", "code review automatique" -> /review-fix skill
• "generer les tests", "auto generate tests", "test suite pour", "genere les tests" -> auto-test-generator agent

Project Management

• "etat du projet", "statut du projet", "quoi de neuf", "project status", "what changed" -> /status skill
• "sante du systeme", "diagnostic", "health check", "check my environment", "verifier ma config" -> /health skill
• "ecris un PRD", "cahier des charges", "specs pour cette feature", "product requirements", "write a PRD" -> /prd skill
• "analyser cette feature", "definir cette idee", "feature analysis", "parlons de cette idee" -> /feature-analyzer skill
• "execute ce design doc", "implement from spec", "on a le spec go", "implement from design doc" -> /feature-pipeline skill

Security & Optimization (extended)

• "audit de securite complet", "full security audit", "vulnerability scan", "verifier les vulnerabilites" -> /security-audit skill
• "optimise tout", "audit de performance", "performance audit", "optimize everything", "benchmark" -> /optimize skill

Infrastructure & DevOps (extended)

• "migrer de", "mise a jour majeure", "upgrade framework", "migrate from", "migrate to" -> /migrate skill
• "configurer GitHub Actions", "pipeline CI", "setup CI/CD", "ajouter CI", "configurer le CI" -> /setup-cicd skill
• "backup ma base", "seed la DB", "schema de base", "database management", "gerer la base" -> /db skill

Deep Analysis & Pre-deploy

• "pense en profondeur", "analyse approfondie", "deep analysis", "think deeply", "ultra think" -> /ultra-think skill
• "validation avant prod", "pre-deploy checklist", "production readiness", "pret pour la prod" -> /pre-deploy skill

Context Management

• "context degradation", "agent handoff", "context budget" -> context-engineering-kit skill

Technical Debt & Data Engineering

• "dette technique", "technical debt", "code health", "audit de dette", "code smell" -> technical-debt-manager agent
• "ETL", "data pipeline", "data processing", "traitement de donnees", "ingestion" -> data-engineer agent

Monorepo & Release

• "monorepo", "workspace", "turborepo", "nx", "pnpm workspaces", "lerna" -> monorepo skill
• "release notes", "changelog", "what's new", "notes de version" -> release-notes skill

Resilience & CLI Testing

• "circuit breaker", "retry pattern", "bulkhead", "resilience pattern", "fallback" -> resilience skill
• "terminal test", "TUI test", "CLI test", "tester le terminal", "interactive CLI" -> terminal-emulator skill

External Services (MCP remote)

• "Stripe", "paiement", "checkout", "ajouter paiement", "add payments", "subscribe" -> Stripe MCP + atum-system:stripe-best-practices skill
• "Netlify", "deploy static" -> Netlify MCP
• "Cloudinary", "upload image", "CDN images" -> Cloudinary MCP
• "docs Microsoft", "Azure" -> Microsoft Learn MCP
• "Linear", "issue tracking", "backlog", "create issue", "creer un ticket" -> Linear plugin
• "Pinecone", "vector store", "semantic search" -> Pinecone plugin + rag-architect skill

Error Tracking & Monitoring (Sentry)

• "error tracking", "monitor errors", "add Sentry", "tracking d'erreurs", "surveiller les erreurs" -> atum-system:sentry-setup-tracing skill
• "Sentry", "setup Sentry", "configurer Sentry", "error monitoring" -> atum-system:sentry-setup-tracing + atum-system:sentry-setup-logging skills
• "AI monitoring", "monitor LLM", "surveiller l'IA", "agent monitoring" -> atum-system:sentry-setup-ai-monitoring skill
• "Sentry metrics", "custom metrics", "metriques" -> atum-system:sentry-setup-metrics skill

Analytics & Feature Flags (PostHog)

• "analytics", "ajouter analytics", "track events", "tracker les evenements" -> atum-system:posthog-instrumentation skill
• "A/B test", "experiment", "feature flag", "flag feature" -> atum-system:flags + atum-system:experiments skills
• "PostHog", "configurer PostHog", "setup analytics" -> atum-system:posthog-instrumentation skill
• "dashboard analytics", "voir les metriques", "insights" -> atum-system:insights skill

Backend-as-a-Service (Firebase / Supabase)

• "Firebase", "add Firebase", "Firebase auth", "authentification Firebase", "Firestore" -> firebase plugin skills
• "Firebase hosting", "deploy Firebase", "Firebase functions" -> firebase plugin skills
• "Supabase", "setup Supabase", "Supabase auth", "Supabase database" -> supabase plugin skills
• "edge function", "Supabase function", "serverless Supabase" -> supabase:deploy_edge_function
• "creer une base de donnees", "setup database", "nouvelle base" -> supabase:create_project or /db skill

Deployment Platforms

• "deploy on Vercel", "deployer sur Vercel", "Vercel", "deploy frontend" -> atum-system:deploy skill
• "deploy on Railway", "Railway", "deployer sur Railway" -> Railway MCP
• "deploy on Render", "Render", "deployer sur Render" -> /deploy skill

Design-to-Code (Figma)

• "implement this design", "from Figma", "Figma URL", "implementer le design" -> atum-system:implement-design skill
• "connect component", "code connect", "connecter composant Figma" -> atum-system:code-connect-components skill
• "design system rules", "regles design system" -> atum-system:create-design-system-rules skill
• figma.com URL detected -> atum-system:implement-design skill

Project Management (Atlassian/Jira)

• "create Jira ticket", "creer un ticket Jira", "Jira issue" -> atum-system:triage-issue skill
• "backlog from spec", "spec to backlog", "creer le backlog" -> atum-system:spec-to-backlog skill
• "meeting notes", "action items from meeting", "taches de la reunion" -> atum-system:capture-tasks-from-meeting-notes skill
• "status report", "rapport d'avancement", "project report" -> atum-system:generate-status-report skill
• "search knowledge", "find in docs", "chercher dans les docs internes" -> atum-system:search-company-knowledge skill

Web Research (Firecrawl)

• "recherche en ligne", "check the web", "search online", "find documentation", "chercher sur le web" -> atum-system:firecrawl-cli skill
• "scrape", "extract from URL", "read this URL", "lire cette page" -> atum-system:firecrawl-cli skill
• "deep research", "investigate", "enqueter", "recherche approfondie en ligne" -> atum-system:firecrawl-cli skill

Static Analysis & Security Scanning

• "static analysis", "SAST", "analyse statique", "scan de code", "semgrep" -> semgrep plugin
• "dependency check", "vulnerable dependency", "CVE check" -> sonatype-guide plugin + supply-chain-risk-auditor skill

Brainstorming & Debugging (Superpowers)

• "brainstorm", "brainstorming", "idees", "explore les options", "let's think" -> atum-system:brainstorming skill
• "debug this", "pourquoi ca marche pas", "ca plante", "investigate this error" -> atum-system:systematic-debugging skill
• "review my PR", "PR review", "review cette PR" -> atum-system:review-pr skill
• "simplify this code", "simplifier ce code", "nettoyer le code" -> atum-system:refactor-cleaner agent

AI/ML & HuggingFace

• "train a model", "fine-tune", "entrainer un modele", "fine-tuner" -> atum-system:hugging-face-model-trainer skill
• "upload to HuggingFace", "publier un modele", "HuggingFace Hub" -> atum-system:hugging-face-cli skill
• "dataset HuggingFace", "creer un dataset", "create dataset" -> atum-system:hugging-face-datasets skill
• "GPU job", "cloud training", "run on HF", "lancer un job" -> atum-system:hugging-face-jobs skill

Interactive Playground

• "create a playground", "interactive explorer", "creer un playground", "outil interactif" -> atum-system:playground skill
• "demo interactive", "visualiser en HTML", "sandbox" -> atum-system:playground skill

Browser Testing (Playwright)

• "test in browser", "tester dans le navigateur", "E2E test", "test bout en bout" -> playwright plugin or atum-system:e2e skill
• "screenshot test", "capture d'ecran test", "visual test" -> atum-system:e2e skill

Documentation Lookup (Context7)

• "check docs for", "documentation de", "how does X work", "comment fonctionne" -> context7 plugin (resolve-library-id then query-docs)
• "latest API docs", "docs a jour", "up-to-date docs" -> context7 plugin

Feature Development

• "guided feature dev", "develop this feature step by step", "developper cette feature" -> atum-system:feature-dev skill
• "explore this codebase", "comprendre ce code", "analyze architecture" -> atum-system:code-explorer agent

Git Workflow

• "commit", "commit and push", "create PR", "creer une PR", "push and PR" -> atum-system:commit-push-pr skill
• "clean branches", "nettoyer les branches" -> atum-system:clean_gone skill

Task Management (Asana)

• "Asana", "tache Asana", "Asana task", "Asana project" -> asana plugin

GitLab

• "GitLab", "merge request", "GitLab CI", "pipeline GitLab" -> gitlab plugin

Laravel

• "Laravel", "artisan", "Eloquent", "blade template" -> laravel-boost plugin

Autonomous Loops

• "ralph loop", "boucle autonome", "autonomous loop", "run continuously" -> atum-system:ralph-loop skill

Browser Testing & Automation

• "test in real browser", "tester dans un vrai navigateur", "browser automation", "automatiser le navigateur" -> agent-browser skill
• "test E2E", "end to end test", "test bout en bout", "playwright test" -> atum-system:e2e skill + e2e-runner agent

Onboarding & Memory

• "bienvenue", "comment ca marche", "je commence", "premier jour", "getting started", "how does this work" -> bienvenue skill
• "souviens-toi", "memoire collective", "sauvegarder cette info", "remember this", "store this decision" -> memoire skill
• "retrouver une info", "chercher dans la memoire", "what did we decide about" -> memoire skill

Project Patterns & Templates

• "skeleton project", "template projet", "project template", "modele de projet" -> project-patterns skill
• "B12 website", "generer un site B12" -> project-patterns skill (B12 pattern)

Image Management

• "image trop grande", "resize image", "redimensionner", "image bloquante" -> image-guard-rules skill (auto-triggered by hooks)

ECC Specialized Agents (auto-triggered by context)

• Go code modified -> atum-system:go-reviewer agent (auto)
• Python code modified -> atum-system:python-reviewer agent (auto)
• Database query/schema work -> atum-system:database-reviewer agent (auto)
• Tests needed -> atum-system:tdd-guide agent (auto, via /tdd command)
• Documentation update needed -> atum-system:doc-updater agent (auto)
• Build fails -> atum-system:build-error-resolver agent (auto)

Superpowers Workflow (auto-triggered)

• Before claiming work is done -> atum-system:verification-before-completion (auto)
• Before writing implementation code -> atum-system:test-driven-development (auto)
• When planning multi-step tasks -> atum-system:writing-plans (auto)
• When facing 2+ independent tasks -> atum-system:fresh-execute (parallel debug mode)

Build & Quality (consolidated from ECC)

• "build error", "erreur de build", "corriger le build", "fix build", "compilation error" -> /build-fix command
• "quality gate", "gate qualite", "check before merge", "verification avant merge" -> /quality-gate command
• "couverture de tests", "test coverage", "code coverage", "rapport de couverture" -> /test-coverage command
• "verifier tout", "verify everything", "verification complete", "check everything works" -> /verify command
• "nettoyer le code mort", "dead code", "unused code", "remove dead code", "refactor clean" -> /refactor-clean command
• "code review", "review de code", "revoir le code" -> /code-review command

Orchestration Multi-Agents (consolidated)

• "plan d'implementation", "write a plan", "planifier l'implementation" -> /plan command
• "orchestrer", "multi-agent", "orchestrate agents" -> /orchestrate command
• "executer en parallele", "parallel execution", "multi execute" -> /multi-execute command
• "plan multi-composants", "multi plan" -> /multi-plan command
• "contexte frais", "fresh context", "sous-taches atomiques", "parallel debug" -> /fresh-execute command
• "executer le plan avec review", "subagent par tache" -> subagent-driven-development skill

Git Workflow (consolidated from commit-commands)

• "commit mes changements", "git commit", "create commit" -> /commit command
• "push et creer une PR", "commit push PR", "ouvrir une PR" -> /commit-push-pr command
• "nettoyer les branches mortes", "clean gone branches", "branches obsoletes" -> /clean_gone command

Plugin Development (consolidated from plugin-dev/hookify)

• "creer un plugin", "scaffold plugin", "new plugin" -> /create-plugin command
• "creer un hook depuis la conversation", "hookify this", "generer un hook" -> /hookify command

Code Review External (consolidated from coderabbit/pr-review-toolkit)

• "coderabbit review", "review avec coderabbit" -> /coderabbit-review command
• "review cette PR", "PR review complete", "review PR" -> /review-pr command

Meta Plugins (auto-triggered — no NLP routing needed)

These plugins are triggered by specific actions, not user intent:

- agent-sdk-dev: triggered when building Claude SDK apps

- claude-code-setup: triggered when setting up Claude Code

- claude-md-management: triggered when managing CLAUDE.md files

- hookify: triggered when creating/managing hooks

- plugin-dev: triggered when creating plugins

- skill-creator: triggered when creating skills

- qodo-skills: auto-loads rules before code tasks

- serena: semantic code tools (auto-invoked by MCP)

- project-architect: triggered for architecture analysis

Composite Request Decomposition

When the user's message contains multiple distinct intents (e.g., "audite la sécurité, optimise les perfs, et crée une PR"), decompose BEFORE routing:

1. Parse — Identify each distinct intent in the message
2. Route independently — Match each intent to its skill/agent separately
3. Sequence — Execute in logical order (e.g., audit before fix, fix before PR)
4. Parallelize when safe — Independent intents can run as parallel agents

Examples

User says	Decomposition
"Audite et déploie"	1. /security-audit → 2. /deploy (sequential: audit must pass first)
"Teste, review et commit"	1. /tdd → 2. /code-review → 3. /commit (sequential chain)
"Fix les tests et update la doc"	1. /build-fix + 2. /update-docs (parallel: independent)
"Crée le composant, ajoute les tests, et fais la PR"	1. implement → 2. /tdd → 3. /commit-push-pr (sequential)

Rule: Never Drop Intents

If only one skill is invoked for a multi-intent request, the routing has failed. After completing each intent, check if remaining intents are pending.

Decision Authority

Make decisions autonomously for:

• File structure and organization
• Naming conventions (follow codebase patterns)
• Error handling strategy
• Test coverage approach
• Package/dependency choices (when standard)

Ask the user for:

• Business logic with multiple valid approaches
• UX/design choices affecting user experience
• Technology stack selection (major choices)
• Breaking changes to public APIs
• Cost-impacting decisions (paid services, scaling)