agnix addresses the critical need for validating AI coding assistant configurations, targeting developers who integrate tools like Claude Code, Copilot, and Cursor. It prevents silent failures caused by syntax errors or misconfigurations in agent skills and instructions, ensuring AI tools function as intended and saving developers significant debugging time.

How It Works

This project provides a rule-based linter and Language Server Protocol (LSP) implementation, built as a Rust workspace with crates like agnix-core for the validation engine and agnix-cli for the command-line interface. It validates various AI agent configuration files (e.g., CLAUDE.md, SKILL.md, hooks, MCP configs) against over 400 rules derived from official specifications, academic research, and practical failure patterns. The core advantage lies in its ability to detect subtle errors that cause AI skills to fail invisibly, offering auto-fix capabilities to resolve issues automatically or with user guidance.

Quick Start & Requirements

Installation is available via npm install -g agnix, brew tap agent-sh/agnix && brew install agnix, or cargo install agnix-cli. Editor extensions are provided for VS Code, JetBrains, Neovim, and Zed. A GitHub Action (agent-sh/agnix@v0) is available for CI integration. An in-browser playground allows immediate testing without installation.

Highlighted Details

• Comprehensive validation with 423 rules covering Claude Code, Codex CLI, OpenCode, Cursor, Copilot, and more.
• Supports a wide range of configuration file types: CLAUDE.md, SKILL.md, AGENTS.md, hooks, MCP configs, and tool-specific files like .cursorrules.
• Features auto-fix capabilities (--fix, --fix-safe, --fix-unsafe) to automatically correct detected issues.
• Integrates seamlessly with major IDEs (VS Code, JetBrains, Neovim, Zed) and supports CI/CD workflows via GitHub Actions.

Maintenance & Community

Contributions are welcomed via CONTRIBUTING.md. Users can report bugs or request new rules through provided links. The project appears actively maintained with continuous addition of new rules and tool support.

Licensing & Compatibility

Licensed under MIT OR Apache-2.0, offering flexibility for both open-source and commercial use.

Limitations & Caveats

While comprehensive, the complexity of AI agent configurations means subtle, context-dependent errors might still require manual review. The README does not explicitly detail unsupported platforms or known bugs.