AgentSkillsCN

agents-md-generator

通过分析项目结构、依赖关系及所用工具，为软件项目生成 AGENTS.md 文件。当用户要求创建 AGENTS.md 文件、为 AI 代理文档化项目，或为代理的使用制定项目规范时，可选用此技能。触发条件包括：“创建 AGENTS.md”、“生成代理说明”、“为该项目撰写代理文档”或“添加 AI 代理文档”等请求。

SKILL.md
--- frontmatter
name: agents-md-generator
description: Generate AGENTS.md files for software projects by analyzing project structure, dependencies, and tooling. Use when asked to create an AGENTS.md file, document a project for AI agents, or establish project conventions for agent consumption. Triggers on requests like "create AGENTS.md", "generate agent instructions", "document this project for agents", or "add AI agent documentation".

AGENTS.md Generator

Generate structured AGENTS.md files by analyzing a project's tools, dependencies, and organization.

Workflow

  1. •Identify project root — Locate package.json, pyproject.toml, Cargo.toml, or similar manifest
  2. •Analyze dependencies — Extract framework, testing, linting, and build tools from manifest
  3. •Scan structure — Map src/, lib/, test/, and other key directories
  4. •Detect quality tools — Identify typecheck, lint, test, and build commands
  5. •Generate AGENTS.md — Output using the template format

Analysis Checklist

Detect Project Type

Manifest FileLanguageCommon Frameworks
package.jsonJavaScript/TypeScriptReact, Next.js, Express, Vite
pyproject.toml / setup.pyPythonFastAPI, Django, Flask
Cargo.tomlRustActix, Axum, Tokio
go.modGoGin, Echo, Fiber
pom.xml / build.gradleJava/KotlinSpring, Quarkus

Identify Quality Tools

Typechecking:

  • •TypeScript: tsc --noEmit
  • •Python: mypy, pyright, ty check
  • •Rust: cargo check

Linting:

  • •JavaScript/TypeScript: eslint, biome
  • •Python: ruff, flake8, pylint
  • •Rust: clippy

Testing:

  • •JavaScript/TypeScript: vitest, jest, mocha
  • •Python: pytest, unittest
  • •Rust: cargo test

Determine Project Classification

  • •Library: Exports modules for other packages (has exports or lib config, no bin)
  • •Application: Runnable entry point (has main, bin, or serves HTTP)
  • •Tool: CLI utility (has bin field or command-line interface)

Output Template

Use the following structure exactly:

markdown
# <Project Name> <Library|Application|Tool>

<1-2 sentence description of what the project does and its purpose>

## Quality Instructions

- **Typecheck**: `<command>` — <brief note if needed>
- **Lint**: `<command>` — <brief note if needed>
- **Test**: `<command>` — <brief note if needed>
- **Build**: `<command>` — <brief note if needed> (if applicable)

## Structure

- `src/` — <purpose>
- `lib/` — <purpose>
- `test/` — <purpose>
<additional key directories only>

## Technical Stack

- **Language**: <primary language and version if relevant>
- **Framework**: <main framework if applicable>
- **Key Libraries**: <2-4 most impactful dependencies>
- **Build Tool**: <bundler or build system>

Generation Guidelines

Be Minimal

  • •List only directories that matter for understanding the codebase
  • •Include only the most impactful 3-5 dependencies
  • •Omit obvious tooling (e.g., don't list typescript if already listing tsc)

Be Specific

  • •Use actual command invocations, not generic descriptions
  • •Reference real paths from the analyzed project
  • •Match the project's actual tooling configuration

Be Concise

  • •One-line descriptions for directories
  • •No redundant explanations
  • •Skip sections that don't apply (e.g., no "Framework" for a pure utility library)

Example Output

markdown
# @acme/data-utils Library

Lightweight data transformation utilities for parsing, validating, and formatting structured data.

## Quality Instructions

- **Typecheck**: `tsc --noEmit`
- **Lint**: `eslint --fix src/`
- **Test**: `vitest run`
- **Build**: `pnpm build`

## Structure

- `src/` — Source modules organized by domain (parsers, validators, formatters)
- `src/__tests__/` — Unit tests colocated with source

## Technical Stack

- **Language**: TypeScript 5.x
- **Key Libraries**: zod (validation), date-fns (date formatting)
- **Build Tool**: tsup