agent-rules-lint

New: SARIF reports, GitHub Action and pre-commit integration. Usage and examples.

Lint AI agent instruction files before they confuse your coding agent.

agent-rules-lint is a small, dependency-free Python CLI that scans files such as AGENTS.md, CLAUDE.md, Cursor rules, and GitHub Copilot instructions for common quality and safety problems.

Why

Recent high-growth developer repos show a clear pattern: teams are adding more agent skills, rule files, memory files, and coding-agent playbooks. The missing piece is a quick quality gate for those instructions.

Bad agent instructions are expensive. They can be too long, vague, contradictory, unsafe, or accidentally include secrets. This tool gives you a fast local and CI check before those rules affect real work.

What It Checks

• Common instruction files:
  • AGENTS.md
  • CLAUDE.md
  • GEMINI.md
  • .cursorrules
  • .cursor/rules/*.md
  • .github/copilot-instructions.md
  • .github/instructions/*.md
• Secret-like values
• Risky shell and Git commands
• Prompt-injection-like language
• Missing title
• Missing purpose, scope, command, or safety guidance
• Vague language
• Files that are too long for practical agent use
• A small set of obvious contradictory rules

Install

Run it without installing:

uvx agent-rules-lint .

Or install it as an isolated CLI:

pipx install agent-rules-lint
agent-rules-lint .

Standard pip installation also works:

python -m pip install agent-rules-lint

For local development:

python -m pip install -e .

Usage

Scan the current repository:

agent-rules-lint .

Return JSON for CI or custom reports:

agent-rules-lint . --format json

Fail CI on warnings:

agent-rules-lint . --warnings-as-errors

Change the long-file threshold:

agent-rules-lint . --max-chars 8000

Example Output

agent-rules-lint checked 1 file(s)
errors=0 warnings=1 info=2

Files:
  - AGENTS.md

Findings:
  [warning] AGENTS.md:42 risky-command: Risky command should include explicit safeguards or approval rules.
  [info] AGENTS.md missing-safety: Consider adding a clear safety section.

CI

Add this to GitHub Actions:

name: Lint agent rules

on:
  pull_request:
  push:
    branches:
      - main

jobs:
  lint-agent-rules:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v7.0.0
      - uses: actions/setup-python@v6.3.0
        with:
          python-version: "3.12"
      - run: pipx run agent-rules-lint . --warnings-as-errors

Design Principles

• No model calls
• No network calls
• No dependencies
• Safe to run in CI
• Findings should be specific enough to act on
• Warnings should improve instruction quality, not enforce one writing style

Current Limitations

• Conflict detection is intentionally conservative and only catches obvious pairs.
• It does not estimate real tokenizer counts; it uses character length as a practical proxy.
• It does not validate whether an instruction is correct for a specific tool.

Development

python -m pip install -e .
python -m unittest discover -s tests -v
python -m agent_rules_lint .

Agent OSS Toolkit

This project is part of a small toolkit for building and launching agent-ready open-source repositories:

• agent-repo-kit: scaffold launch-ready, AI-agent-friendly repositories
• oss-launch-check: audit whether a repository is ready to launch as open source
• repo-context-card: generate compact repository context cards for coding agents
• agent-rules-lint: lint AGENTS.md, CLAUDE.md, Cursor rules, and Copilot instructions
• awesome-ai-agents-zh: Chinese AI Agents / MCP / AI DevTools directory

Contributing

See CONTRIBUTING.md.

License

MIT