meta-agent-creator

name	meta-agent-creator
description	Create Pi agent definitions with proper frontmatter, model tiers, and tool permissions. Use when building specialized agents for code review, documentation, security, research, etc.
license	MIT
compatibility	pi
metadata	{"audience":"developers","category":"meta"}

Meta Agent Creator

Create specialized Pi agent personas using pi-agents.

What I Do

• Create agent markdown files with proper frontmatter
• Configure model tier selection
• Set up extension loading for spawned agents
• Guide tool permissions and skill injection

When to Use Me

Use this skill when:

• Creating specialized agents (code review, docs, security, etc.)
• Setting up agents for parallel work via spawn_agent
• Configuring agent tool access and models

Do NOT use for:

• Creating skills (use meta-skill-creator)
• Creating extensions (use meta-extension-creator)
• Creating prompt templates (use meta-command-creator)

Model Tiers

Model tiers are defined in .pi/agents.json:

{
  "models": {
    "lite": "<your-fast-model>",
    "medium": "<your-balanced-model>",
    "high": "<your-best-model>",
    "oracle": "gpt-5.3-codex"
  }
}

Use tier names in agent frontmatter. pi-agents resolves them to actual model strings at spawn time.

Tier	Use Case
lite	Fast tasks: scouting, context gathering
medium	Balanced: code review, implementation
high	Complex: planning, orchestration
oracle	Deep reasoning, architecture

Quick Start

Basic Agent

Create .pi/agents/<name>.md:

---
name: reviewer
description: Reviews code for best practices and security
model: medium
extensions: pi-agents, pi-question
---

You are a senior code reviewer. Focus on:
- Code quality and best practices
- Security vulnerabilities
- Performance implications

Provide detailed feedback in markdown format.

Agent with Skills

---
name: git-helper
description: Assists with git operations and commit messages
model: lite
extensions: pi-agents, pi-question
skill: git
---

You help with git operations. Follow conventional commits.
Use the git skill for best practices.

Agent Locations

Scope	Path
Project	.pi/agents/{name}.md
Package	Via agentPaths in .pi/agents.json

Frontmatter Options

Option	Type	Description
name	string	Agent identifier
description	string	When to use this agent
model	string	Tier name (lite, medium, high) or raw model string
thinking	string	Thinking level: off, minimal, low, medium, high, xhigh
tools	string	Comma-separated: read, grep, find, ls, bash, write, edit
extensions	string	Comma-separated pi package names for spawn_agent to load
skill	string	Comma-separated skills to inject into system prompt

Invoking Agents

spawn_agent (tmux — visible work)

Use spawn_agent when the user should see the agent's work:

spawn_agent({ agent: "reviewer", task: "Review the auth module" })

Parallel visible work:

spawn_agent({ agent: "reviewer", task: "Review frontend" })
spawn_agent({ agent: "reviewer", task: "Review backend" })

Common Patterns

Pattern	Tools	Model Tier	Use Case
Scout	read, grep, find, ls	lite	Fast codebase analysis
Reviewer	read, grep, find, ls	medium	Code review, no changes
Builder	all	medium	Implementation
Planner	read, grep, find, ls	high	Complex planning
Researcher	read, grep, find, bash	lite	Information gathering

Tips

1. Minimal tools: Only grant tools the agent actually needs
2. Use skills: Inject domain knowledge via skill: frontmatter
3. Lite for speed: Use lite tier for fast, focused tasks
4. Medium for quality: Use medium tier for implementation
5. High for complexity: Use high tier for planning/orchestration
6. Extensions list: Include pi-agents so spawned agents can spawn sub-agents too

Detailed References

• Configuration — All frontmatter options and model tiers
• Examples — Complete agent examples
• Permissions — Tool permissions details