メニュー
Generate AGENTS.md for cross-agent portability (Claude Code, Cursor, Codex, Gemini CLI). Tool-agnostic repo context document, distinct from CLAUDE.md.
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。
npx skills add https://github.com/UitbreidenOS/UitKit --skill agents-mdコマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。
ローカルで確認しますか？SkillsMP が現在取得できるファイルをダウンロードできます。
SOC 職業分類に基づく
SKILL.md を表示中
| name | agents-md |
| description | Generate AGENTS.md for cross-agent portability (Claude Code, Cursor, Codex, Gemini CLI). Tool-agnostic repo context document, distinct from CLAUDE.md. |
| CLAUDE.md | AGENTS.md |
|---|---|
| Claude Code-specific | Works with any AI coding tool |
| Can reference hooks, MCP, slash commands | Tool-agnostic — no Claude features |
| Can be verbose (loaded once per session) | Should be concise (<400 lines) |
| Project + Claude behavior rules | Project conventions + agent safety rules |
| Owned by Claude Code users | Owned by any AI-assisted team |
The two files can coexist. AGENTS.md is the superset of what all tools share; CLAUDE.md adds Claude Code-specific extensions on top.
Every AGENTS.md must contain these sections in order:
1. Project Overview Two to four sentences. What the project does, who it serves, what problem it solves. No marketing language.
2. Tech Stack Bulleted list: language + version, framework + version, major libraries, database, hosting/infrastructure. Only what is actually in the repo — no aspirational additions.
3. Key Conventions The rules any developer (or AI agent) must follow to produce acceptable code. Pull from:
.eslintrc, pyproject.toml, rubocop.yml)Include: naming conventions, file organization, patterns to use, patterns to avoid.
4. Agent Behavior Rules Instructions specifically for AI agents:
5. File Safety Map A table classifying paths by risk:
| Path / Pattern | Status | Notes |
|---|---|---|
src/, lib/, app/ | SAFE | Feature code — normal editing |
tests/, spec/, __tests__/ | SAFE | Tests — modify freely |
docs/ | SAFE | Documentation |
prisma/migrations/, db/migrate/ | CAUTION | Run only with approval |
src/auth/, src/payments/ | CAUTION | Security-sensitive — flag before changing |
.env, *.secret, credentials.* | DANGEROUS | Never read or modify |
Dockerfile, .github/workflows/ | DANGEROUS | Infrastructure — approval required |
dist/, build/, .next/ | DANGEROUS | Generated — do not edit directly |
6. Test Commands
Exact commands copied from package.json scripts, Makefile, or CI config. No paraphrasing.
Test: npm test
Lint: npm run lint
Typecheck: npm run typecheck
Format: npm run format
PreToolUse, PostToolUse, Stop)These belong only in CLAUDE.md.
Use the prompts/task-specific/agents-md-generator.md prompt for a fully automated generation pass. The skill here covers the format; the prompt covers the invocation.
Project: Next.js 14 SaaS app with Prisma + PostgreSQL, deployed on Vercel.
Generated AGENTS.md excerpt:
# AGENTS.md
## Project Overview
A multi-tenant SaaS dashboard for tracking software delivery metrics.
Built for engineering teams of 5–50 people. Core value: answering
"how fast are we shipping and where are we blocked" in under 30 seconds.
## Tech Stack
- TypeScript 5.4
- Next.js 14 (App Router)
- Prisma 5 + PostgreSQL 16
- Tailwind CSS 3.4
- Vercel (hosting + Edge Functions)
- Vitest (unit), Playwright (e2e)
## Key Conventions
- All components: PascalCase, one component per file
- All utilities: camelCase functions, kebab-case filenames
- No barrel exports (no index.ts re-exports)
- Server Components by default; add 'use client' only when necessary
- Database queries only in `src/lib/db/` — never in components or API routes directly
- No `any` types — use `unknown` + type guard if the shape is uncertain
## Agent Behavior Rules
Safe to run autonomously: `pnpm test`, `pnpm lint`, `pnpm typecheck`, `pnpm format`
Require confirmation: `prisma migrate deploy`, `vercel --prod`, any `DROP` SQL
Ask before creating new files in: `src/lib/db/`, `prisma/migrations/`
Do not commit autonomously — always present changes for review first
## File Safety Map
| Path | Status | Notes |
|---|---|---|
| `src/components/` | SAFE | |
| `src/app/` | SAFE | |
| `src/lib/` | CAUTION | Core business logic |
| `prisma/migrations/` | DANGEROUS | Never modify existing migrations |
| `.env*` | DANGEROUS | Never read or write |
| `.github/` | DANGEROUS | CI/CD — approval required |
## Test Commands
pnpm test # Vitest unit tests
pnpm test:e2e # Playwright end-to-end
pnpm lint # ESLint
pnpm typecheck # tsc --noEmit