Every AI coding tool wants a file describing your project. Claude Code reads CLAUDE.md. Cursor reads .cursor/rules. Then there's AGENTS.md, GEMINI.md, Copilot instructions, Aider's CONVENTIONS.md, Windsurf, Cline, Continue, Kilo Code, Trae, Junie, Warp...

Same content, a dozen names and formats. Maintaining them by hand is busywork, and the second your stack changes they're stale.

One scan, every file

npx @horiastanxd/claude-init

It analyzes the repo once - language, framework, package manager, scripts, file tree, env vars from .env.example, conventions (strict mode, linter, formatter), git info - and writes all 13 files.

No API key, no network. 100% local.

Stop the rot with check

The real problem isn't generating once - it's drift. So there's a check mode:

npx @horiastanxd/claude-init check

It regenerates in memory and diffs against what's on disk, exiting non-zero on drift. Drop it in CI or a pre-commit hook:

- run: npx @horiastanxd/claude-init check

Also an MCP server

claude mcp add claude-init -- npx @horiastanxd/claude-init --mcp

Now an agent can analyze a repo and write its own context files.

Details

• Languages: JS/TS (npm/pnpm/yarn/bun), Python (pip/uv/poetry), Rust, Go, Java (Maven/Gradle), Ruby, PHP
• --recurse for monorepos (npm/pnpm workspaces)
• TypeScript, MIT licensed

Repo: https://github.com/horiastanxd/claude-init

What tool or format should it support next?