Reflint — Git skill for Claude Code

reflint


One of three zero-dependency linters for AI-agent repos. To run all three in a single pass, with one report and one exit code, use **[tenken](https://github.com/hyuga611/tenken)** — `npx @hyuga/tenken`.

[](https://www.npmjs.com/package/@hyuga/reflint) [](LICENSE) [](package.json)

**In any language, fail the PR when your `AGENTS.md`, `llms.txt`, or `CLAUDE.md` points at a command, script, or path that no longer exists.** `reflint` is a *reference-integrity* linter for agent config files. It doesn't grade style or prose — it checks whether the references are **real**: the scripts, paths, and files your config tells an AI agent to use. Zero-dependency, language-agnostic, runs in CI on every PR.

**`AGENTS.md` や `llms.txt` の中の「もう解決できない参照」を、どのスタックでも毎PRで落とす。** AIエージェント向け設定ファイルの **参照整合性 (reference integrity)** を CI で検証するリンタ。表記や文体ではなく、AIに渡す **"嘘の指示" そのもの** ── 存在しないコマンド・スクリプト・パス ── を落とす。依存ゼロ・言語非依存。

---

Why / なぜ

Agent config rots silently. When it tells the agent to run a script that was renamed, or points at a path that was deleted, the agent trusts it and breaks things. `reflint` fails CI before that happens. It checks *facts*, not prose — so it works in any stack (Go, Rust, Python, Ruby, JS…), and it treats `llms.txt` as a first-class target, which the rest of the ecosystem's format validators don't check for real references.

Use as a GitHub Action / CIで使う（定着の本体）

# .github/workflows/reflint.yml
name: reflint
on: [push, pull_request]
jobs:
  reflint:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
        with: { fetch-depth: 0 }      # needed for the "new in this PR" diff
      - uses: hyuga611/reflint@v0     # auto-detects AGENTS.md / llms