agent-file-structure（Agent 文件结构）

Agent File Structure Skill

Purpose

Defines the required structure and best practices for creating agent definition files in .github/agents/*.agent.md.

When to Use

• When creating a new agent definition file
• When reviewing or refactoring existing agent definitions
• When ensuring consistency across the agent ecosystem

Required Structure

All agents must follow this structure:

---
description: Brief, specific description (≤100 chars)
name: Workflow Engineer (coding agent)
model: <model name>
---

# Agent Name Agent

You are the **Agent Name** agent for this project...

## Your Goal
Single, clear goal statement.

## Boundaries
✅ Always Do: ...
⚠️ Ask First: ...
🚫 Never Do: ...

## Context to Read
- Relevant docs with links

## Workflow
Step-by-step numbered approach

## Output
What this agent produces

Key Principles

• Specific over general - "Write unit tests for React components" beats "Help with testing"
• Commands over descriptions - Include exact commands: npm test, dotnet build
• Examples over explanations - Show real code examples, not abstract descriptions
• Boundaries first - Clear rules prevent mistakes

Frontmatter Requirements

The YAML frontmatter at the top of each agent file must include:

• description: Brief description (100 characters or less) that explains the agent's purpose
• name: The agent's display name (include "(coding agent)" or local agent type)
• model: VS Code agents only — The language model assigned to this agent (must exist in docs/ai-model-reference.md)

⚠️ **Coding agents (*-coding-agent.agent.md) must NOT include model: in frontmatter.** The model: property is not supported on GitHub.com coding agents and causes a hard CAPIError: 400 The requested model is not supported error, preventing the agent from running. VS Code agents (without the -coding-agent suffix) should always include model: for LLM selection. See docs/ai-model-reference.md for details.

Section Guidelines

Your Goal

• One clear sentence describing what the agent accomplishes
• Focus on outcomes, not process
• Example: "Implement features and tests according to specifications"

Boundaries

• ✅ Always Do: Mandatory actions the agent must take (use specific commands)
• ⚠️ Ask First: Situations requiring maintainer approval before proceeding
• 🚫 Never Do: Actions that are explicitly forbidden or outside agent scope

Context to Read

• List of documentation files the agent should review before starting work
• Use relative paths with links (e.g., [docs/spec.md](../../docs/spec.md))

Workflow

• Numbered steps in logical sequence
• Include exact commands where applicable
• Specify decision points and branching logic

Output

• List of artifacts the agent produces
• Include file locations and formats
• Clarify deliverables vs intermediate work

Best Practices

• Be specific: Include exact file paths, command syntax, and expected outputs
• Use examples: Show don't tell - include code snippets and command examples
• Keep it actionable: Every instruction should be something the agent can execute
• Avoid ambiguity: Use precise language and avoid vague terms like "usually" or "might"