---

name: vg-master-orchestrator description: Master Orchestrator for autonomous multi-agent coordination. Use when tasks require multiple technology experts, complex workflows (TDD pattern addition, security audits), intelligent routing, or explicit orchestration requests. Coordinates 17 technology experts with real-time progress reporting. version: 2.0.0 allowed-tools: [Read, Write, Edit, Bash, Grep, Glob, Task, SlashCommand, Skill]

Master Orchestrator - Execution Instructions (v2.0.0)

Purpose

Execute autonomous multi-agent coordination for Vigil Guard v2.0.0 tasks. Analyze, classify, route to technology experts, and synthesize results.

Quick Reference

Architecture: 3-Branch Parallel Detection (24 nodes) Available experts: 17 technology experts in .claude/agents/ Full details: See CLAUDE.md section "Technology Expert Agents (v3.0)"

Technology Experts (17)

Expert	Technology	Use For
orchestrator	Coordination	Multi-expert task routing
n8n-expert	n8n	Workflows, nodes, webhooks
react-expert	React + Vite	Components, hooks, UI
express-expert	Express.js	REST APIs, middleware
vitest-expert	Vitest	Testing, TDD, fixtures
clickhouse-expert	ClickHouse	Analytics SQL, schema
docker-expert	Docker	Containers, compose (11 services)
presidio-expert	MS Presidio	PII detection, NLP
security-expert	Security	OWASP, auth, vulnerabilities
git-expert	Git	Version control, commits
python-expert	Python	Flask, FastAPI
tailwind-expert	Tailwind CSS	Styling
kubernetes-expert	Kubernetes	Cluster ops, pods, deployments
helm-expert	Helm	Charts, releases, templating
nats-expert	NATS	Messaging, JetStream (future)
redis-expert	Redis	Caching, rate limiting (future)
code-audit-expert	Code Auditing	Quality assessment, tech debt

v2.0.0 Architecture Context

3-Branch Parallel Detection System

Input → Validate → Config Load
         ↓
    3-Branch Executor (PARALLEL):
    ├─ Branch A: Heuristics (:5005) - 30% weight
    ├─ Branch B: Semantic (:5006) - 35% weight
    └─ Branch C: LLM Guard (:8000) - 35% weight
         ↓
    Arbiter v2 (weighted fusion)
         ↓
    Decision → PII Redaction → ClickHouse Log

Key Services (11 Docker containers)

• vigil-heuristics (5005) - Pattern detection
• vigil-semantic-service (5006) - Embedding similarity
• vigil-presidio-pii (5001) - PII detection
• vigil-language-detector (5002) - Language detection
• vigil-n8n (5678) - Workflow engine
• vigil-clickhouse (8123) - Analytics DB
• vigil-grafana (3001) - Monitoring
• web-ui-backend (8787) - Config API
• web-ui-frontend (80) - React UI
• vigil-prompt-guard-api (8000) - LLM safety
• vigil-caddy (80/443) - Reverse proxy

Execution Steps

1. Classify Task (Lightweight)

Determine strategy WITHOUT reading files:

• Single: One expert needed
• Parallel: Multiple independent experts
• Sequential: Experts depend on previous results
• Workflow: Matches pre-configured template

Workflows:

• PATTERN_ADDITION (TDD): vitest-expert → n8n-expert → vitest-expert
• SECURITY_AUDIT (Parallel): security-expert + vitest-expert + docker-expert
• PII_ENTITY_ADDITION: presidio-expert → express-expert → react-expert → vitest-expert

2. Execute with Progress Reporting

Format:

🎯 Task: [description]
🎭 Strategy: [single|parallel|sequential|workflow]

🤖 Expert: [name]-expert
├─ ▶️  Action: [action]
├─ 📝 [progress]
└─ ✅ Completed (X.Xs)

✨ Task Completed

3. Expert Invocation Methods

Option A - Use existing skill (FASTEST):

• vigil-testing-e2e for testing tasks
• n8n-vigil-workflow for workflow/pattern tasks
• presidio-pii-specialist for PII tasks
• react-tailwind-vigil-ui for UI tasks
• docker-vigil-orchestration for container tasks
• etc.

Option B - Task tool with technology expert:

Task(
  prompt="You are n8n-expert. [task description]",
  subagent_type="n8n-expert"
)

Option C - Direct implementation:

• Use Read/Write/Edit/Bash tools directly
• Only when no skill/expert available

4. Result Synthesis

═══════════════════════════════════════
✨ Task Completed in X.Xs

📋 Summary: [what was done]

🤝 Coordinated N experts:
   • n8n-expert
   • vitest-expert

💡 Next Steps:
   1. [action 1]
   2. [action 2]

Critical Rules

DO:

• ✅ Use existing skills when available (performance)
• ✅ Show real-time progress with emoji indicators
• ✅ Keep skill operations lightweight
• ✅ Reference v2.0.0 architecture (3-branch parallel, 24 nodes)
• ✅ Synthesize results concisely

DON'T:

• ❌ Reference old vg-* agents (REMOVED in v3.0)
• ❌ Reference 40-node sequential pipeline (NOW 24-node parallel)
• ❌ Reference rules.config.json (MERGED into unified_config.json)
• ❌ Read .claude/agents/*/AGENT.md unless truly needed
• ❌ Just show documentation (EXECUTE the work)
• ❌ Load unnecessary context

Token Optimization

Problem: This skill + CLAUDE.md can cause session hangs due to token overload.

Solution:

1. Classify tasks WITHOUT reading files
2. Use existing skills (they're already loaded)
3. Only read AGENT.md if truly needed
4. Keep progress reporting concise

Error Handling

If expert fails:

1. Report error with expert name
2. Retry once
3. Suggest fallback expert
4. Continue with remaining work if possible

Integration

Slash command: /vg-orchestrate [task] → invokes this skill Direct: User can also invoke via @vg-master-orchestrator or Skill tool

Version History

• v2.0.0 (2025-12): Updated for 3-branch architecture, 17 technology experts
• v2.0.1 (legacy): 10 vg-* agents (DEPRECATED)

Ready to orchestrate!