ai-kit

One setup for every AI coding agent.

One command gives any project the same rules, docs and proven prompts, so Claude Code, Gemini / Antigravity, Codex and Cursor all work the same careful way, from the first session.

Quick start

Inside your project folder:

npx github:Hazman97/ai-kit init

Then tell your AI agent:

follow .ai/prompts/project-setup.md

That's it: your agent fills docs/PROJECT.md from the code, and every agent now follows the same rules. No install and no dependencies; Node.js 18+ is enough. Full guide with examples: hazman97.github.io/ai-kit

What you get

your-project/
├── AGENTS.md            ← the rules every agent follows (one source for all of them)
├── CLAUDE.md            ← "@AGENTS.md" (Claude Code reads AGENTS.md through it)
├── GEMINI.md            ← "@./AGENTS.md" (Gemini CLI / Antigravity too)
├── LEARNINGS.md         ← mistakes worth remembering; agents add entries only with your OK
├── docs/
│   └── PROJECT.md       ← the one base doc: what it is, stack, commands (filled for you), structure
└── .ai/
    ├── ai-kit.json      ← version + what was detected
    └── prompts/         ← tools for bigger jobs, used when you ask for them
        ├── project-setup.md   fill docs/PROJECT.md from the code
        ├── docs-update.md     sync docs with code (update only, verify every claim)
        ├── docs-migrate.md    merge docs the project already had into this layout (once)
        ├── frontend-setup.md  write docs/FRONTEND.md (design tokens, Tailwind rules, phone rules)
        ├── api-setup.md       write docs/API.md (routes, auth, payloads, errors)
        ├── database-setup.md  write docs/DATABASE.md (schema, storage, access rules, migrations)
        ├── ui-task.md         hand UI work to another agent: logic first, look second, screenshots as proof
        ├── audit.md           a production-safe audit: static + API contract + running the app
        └── security-audit.md  Cloudflare projects only: Cloudflare's security-audit skill + a Cloudflare checklist

The ideas behind it

• Few docs, always true. One base doc (docs/PROJECT.md). Other docs (API, DATABASE, FRONTEND, OPS) are created only when the project needs them. The Docs Index in AGENTS.md lists only files that exist, so agents never chase a doc that is not there.
• Rules that hold up: verify before saying "done", fix root causes, never commit or run destructive commands without asking, secrets never in code, phone-width and light/dark checks for every screen.
• Your text is yours. ai-kit only changes its own marked blocks. Everything else (your Project notes, LEARNINGS.md, docs/PROJECT.md, prompts you edited) is never overwritten.

Commands

Run them inside the project folder (or pass --dir <path>). A folder given any other way (ai-kit update ../other) or an unknown option is refused before anything is written.

Command	What it does
npx github:Hazman97/ai-kit init	Detects the stack in the root and every sub-folder with its own manifest (a frontend/ + backend/ repo, apps/* workspaces, a Python or Go API next to a React app…), labels each part frontend / backend / fullstack / mobile, and adds the files above. Existing files are kept; .gitignore gets a small ai-kit block so agent scratch files (screenshots, browser profiles) are never committed.
npx github:Hazman97/ai-kit init (again, or on a project that already has an AGENTS.md)	Brings it to the latest version: an ai-kit AGENTS.md gets the newest rules; any other AGENTS.md / CLAUDE.md / GEMINI.md is backed up to .ai/backup/, gets the rules and the import line, and keeps its own sections and Docs Index texts. --no-merge leaves them untouched.
npx github:Hazman97/ai-kit update	Newer rules and prompts from this repo. Your notes and your own prompts are not touched.
npx github:Hazman97/ai-kit doctor	Checks: docs referenced but missing, unfilled placeholders, old doc names, a stale Docs Index, agents that do not read AGENTS.md, a long LEARNINGS.md, and .env files committed to git. Exit code 1 on errors (handy in CI).
npx github:Hazman97/ai-kit doctor --fix	Also regenerates the Docs Index.
npx github:Hazman97/ai-kit update --security-skill	Cloudflare projects only: installs Cloudflare's security-audit skill into this project for its agents (see below). Any other project: skipped, nothing installed.

Add --dry-run to any command to see what would change without changing anything.

After init

Once per project (later sessions read these docs on their own, and if you forget, the agent offers to do it when it finds the doc empty). Open the project with your agent and say:

1. follow .ai/prompts/project-setup.md → it fills docs/PROJECT.md from the code. The project already has its own docs (3 or more)? follow .ai/prompts/docs-migrate.md instead: it merges what is still true into the ai-kit docs, checked against the code, and asks before deleting the old files.
2. For a project with a UI: follow .ai/prompts/frontend-setup.md → it writes docs/FRONTEND.md. A backend with many routes or a real schema: api-setup.md → docs/API.md, database-setup.md → docs/DATABASE.md (only when the project needs them).
3. Commit the new files.

Later: follow .ai/prompts/audit.md before a release, follow .ai/prompts/ui-task.md to hand a screen to another agent, and "update the docs" whenever the code moved on.

여기까지가 README 앞부분입니다. 나머지는 원문에서 읽을 수 있습니다. README 전문 보기 ↗