pi-auto-agents

Intelligent task orchestrator that breaks down complex requests using specialized agents

The pi-auto-agents skill automatically detects complex coding and development tasks and orchestrates them across specialized sub-agents for thorough analysis, planning, implementation, testing, and review.

Important: This skill requires the pi-subagents extension to work. It depends on the subagent and subagent_status tools provided by pi-subagents.

Features

• Intelligent Task Routing: Analyzes requests to determine when multi-agent coordination is needed
• Specialized Agents: Pre-built agents for different phases of development
• Workflow Orchestration: Supports sequential chains and parallel execution
• Mandatory Testing & Review: Enforces testing after implementation and quality review
• Progress Tracking: Maintains detailed progress documentation
• Data Flow: Automatic passing of context between agents using {previous} template variables

Available Agents

• scout: Fast codebase exploration and analysis (lightweight and quick)
• context-builder: Builds comprehensive context and meta-prompts from requirements
• planner: Creates detailed implementation plans and task breakdowns
• worker: General-purpose implementation and coding tasks
• researcher: Conducts web research and gathers external information
• reviewer: Code review, validation, and quality assurance

Installation

pi install npm:pi-auto-agents

This package includes the skill definition at skills/auto-agents/SKILL.md and will be automatically discovered by Pi.

How It Works

The orchestration follows a structured workflow:

1. ANALYZE - Thoroughly understand the user's request
2. PLAN - Create an optimal workflow with verification steps
3. EXECUTE - Run chains or parallel agent tasks using the subagent tool
4. TEST & DEBUG - Always test code after implementation and fix issues
5. REVIEW - Perform quality assurance with the reviewer agent
6. VERIFY - Confirm the solution works end-to-end
7. SYNTHESIZE - Provide clear summary of results

Usage

Basic Chain (Recommended for Development Tasks)

{
  "chain": [
    {"agent": "scout"},
    {"agent": "context-builder"},
    {"agent": "planner"},
    {"agent": "worker"},
    {"agent": "reviewer"}
  ]
}

With Custom Tasks and Parallel Execution

{
  "chain": [
    {"agent": "scout", "task": "Analyze current codebase structure"},
    {
      "parallel": [
        {"agent": "worker", "task": "Implement frontend changes"},
        {"agent": "worker", "task": "Update backend API"}
      ]
    },
    {"agent": "reviewer"}
  ]
}

Single Agent Call

{
  "agent": "planner",
  "task": "Create a detailed plan for adding user authentication"
}

Monitoring Async Runs

{
  "agent": "subagent_status",
  "id": "run-id-here"
}

Testing & Debugging Policy

MANDATORY for all implementation tasks:

• After any worker phase, immediately test the changes
• Use bash tool to run tests and verify functionality
• Create reproduction test cases for bugs
• Document all test results in progress.md
• Only proceed to review after tests pass

Dependencies

This skill is heavily dependent on pi-subagents — it provides the core subagent and subagent_status tools that make multi-agent orchestration possible.

Without pi-subagents, this skill will not function.

Acknowledgments

Thank you to the creators and maintainers of:

• pi-subagents — for building the powerful sub-agent infrastructure this skill depends on
• Pi Framework (@mariozechner/pi-coding-agent) — for creating such an elegant and extensible coding agent platform

Your work made this orchestration layer possible. 🙏

Project Structure

.
├── skills/
│   └── auto-agents/
│       └── SKILL.md          # Full skill documentation
├── README.md                 # This file
├── package.json
└── install.mjs

Links

• Skill Documentation
• Repository
• Pi Framework

---

Built for the Pi coding agent harness.