← Back to Discover

agentchecker

Command-line tool that compares the instruction files of AI coding assistants in a project, reports conflicting tech-stack rules, and offers to fix them.

Command-line toolsTesting & QA
Visit website ↗

What it does

agentchecker addresses the case where different AI coding tools read different instruction files in the same repository, such as .cursorrules, CLAUDE.md, Copilot instructions and AGENTS.md, and gradually contradict each other about which package manager, linter or formatter to use. Run with npx in the project root, it scans the project files and optionally your global tool configs, extracts those tech-stack choices and prints a terminal report of the contradictions. It can then apply fixes interactively to local project files only, since global configs stay read-only, and it can create an AGENTS.md or link tool-specific files to it. A check-only mode exits with an error when contradictions exist, which suits pre-commit hooks, and a GitHub Action covers CI. The supported files cover Cursor, Claude Code, GitHub Copilot, Codex, Antigravity, OpenCode, Windsurf, Roo Code and Cline, and Aider.

Capabilities

  • Scans project and optional global instruction files for several AI coding tools
  • Extracts package manager, linter, formatter and test runner choices
  • Interactive fixes that only touch local project files
  • Check-only mode for pre-commit hooks and a GitHub Action for CI
  • Can merge instruction files into AGENTS.md and symlink the others to it

When to use it

  • Your repo has separate rule files for Cursor, Claude Code, Copilot or others and they may disagree
  • You want CI or a pre-commit hook to block conflicting agent instructions
  • You are consolidating several agent instruction files into one AGENTS.md