Pi

bunx @every-env/compound-plugin install compound-engineering --to pi

~/.pi/agent/
├── prompts/
│   └── *.md              # Command prompts
├── skills/
│   ├── agent-name/
│   │   └── SKILL.md      # Agent skills
│   └── existing-skill/   # Pass-through skills
├── extensions/
│   └── compound-engineering-compat.ts  # Compatibility extension
├── compound-engineering/
│   └── mcporter.json     # MCPorter config for MCP access
└── AGENTS.md             # Pi tool mapping notes

---
description: Turn feature ideas into detailed plans
argument-hint: <feature description>
---

You are a planning specialist. Run subagent with agent="repo-research-analyst" and task="research the feature".

## Pi + MCPorter note
For MCP access in Pi, use MCPorter via the generated tools:
- `mcporter_list` to inspect available MCP tools
- `mcporter_call` to invoke a tool

---
name: plan-specialist
description: Planning agent for complex features
---

## Capabilities
- Break down complex features
- Identify dependencies
- Create detailed plans

You are a planning specialist...

- Task repo-research-analyst(feature_description)
+ Run subagent with agent="repo-research-analyst" and task="feature_description".

- Run /workflows:plan
+ Run /workflows-plan

- Use AskUserQuestion to prompt the user
+ Use ask_user_question to prompt the user

- Use TodoWrite to track tasks
+ Use file-based todos (todos/ + /skill:file-todos)

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem"],
      "env": {
        "ALLOWED_PATHS": "/Users/you/projects"
      }
    },
    "remote-api": {
      "baseUrl": "https://api.example.com/mcp",
      "headers": {
        "Authorization": "Bearer ${TOKEN}"
      }
    }
  }
}

// ~/.pi/agent/extensions/compound-engineering-compat.ts

// Provides compatibility shims for:
// - Claude Task syntax → Pi subagent calls
// - Tool name mappings
// - MCP access via MCPorter

<!-- BEGIN COMPOUND PI TOOL MAP -->
## Compound Engineering (Pi compatibility)

This block is managed by compound-plugin.

Compatibility notes:
- Claude Task(agent, args) maps to the subagent extension tool
- For parallel agent runs, batch multiple subagent calls with multi_tool_use.parallel
- AskUserQuestion maps to the ask_user_question extension tool
- MCP access uses MCPorter via mcporter_list and mcporter_call extension tools
- MCPorter config path: .pi/compound-engineering/mcporter.json (project) or ~/.pi/agent/compound-engineering/mcporter.json (global)
<!-- END COMPOUND PI TOOL MAP -->

{
  "name": "workflows:plan",
  "description": "Create plan",
  "body": "Use MCP filesystem tools to read the codebase. Task repo-research-analyst(analyze)."
}

---
description: Create plan
argument-hint: 
---

Use MCP filesystem tools to read the codebase. Run subagent with agent="repo-research-analyst" and task="analyze".

## Pi + MCPorter note
For MCP access in Pi, use MCPorter via the generated tools:
- `mcporter_list` to inspect available MCP tools
- `mcporter_call` to invoke a tool

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem"]
    }
  }
}

bunx @every-env/compound-plugin sync --target pi