Agent Orchestrator Skill

Purpose

Centralized orchestration for delegating exploration and research tasks to Haiku-based sub-agents, enabling 50%+ token reduction and 56%+ cost savings per workflow.

When to Use

• •Before /spec:plan: Delegate codebase exploration and security analysis
• •Before /spec:specify: Find similar implementations and patterns
• •Before /spec:tasks: Analyze dependencies and task ordering
• •Before /code: Run security pre-checks

Usage

pnpm skill:run agent-orchestrator orchestrate '{
  "task_type": "research",
  "context": {
    "query": "How does authentication work?",
    "focus_areas": ["auth", "session"],
    "depth": "deep"
  }
}'

Response Format

Returns structured JSON with findings from sub-agents:

{
  "success": true,
  "agent_used": "research",
  "findings": { ... },
  "tokens_used": 48000,
  "confidence": "high"
}

Available Agents

• •research: Codebase exploration, pattern discovery, external documentation
• •security: Vulnerability scanning, RLS gap analysis, security recommendations
• •test: Test generation (stub for future integration)

Parallel Execution

Use task_type: "multi" to run multiple agents in parallel:

{
  "task_type": "multi",
  "agents": ["research", "security"],
  "context": { ... }
}

Fallback

Falls back to main agent if delegation fails. Check response.success before using findings.