Summary

AIGX provides an open, benchmark-validated context format for AI coding agents, enabling them to understand codebase conventions and behavior. It centralizes rules and per-file indexing in a .aigx/ directory, allowing agents to act like senior engineers without modifying source code.

How It Works

AIGX defines a codebase's AI-agent rules, boundaries, and pitfalls within a .aigx/ directory, forming a "genome." A files.aigx index maps specific files to relevant constraints (forbidden imports, gotchas). This per-file addressability ensures agents retrieve precise binding information at the edit site, improving accuracy. The format is tool-agnostic and avoids source code injection, a design validated as superior in controlled benchmarks.

Quick Start & Requirements

• Installation: Scaffold a genome using npm create aigx or npx create-aigx. Reference implementations are available via npm (@aigx/cli), PyPI (aigx), and crates.io (aigx). Zero-install CLI commands like npx @aigx/cli lint are supported.
• Prerequisites: Standard Node.js, Python, or Rust environments are sufficient.
• Editor Support: VS Code, Cursor, and Windsurf integrate via the "AIGX Language Support" extension.
• Links: Quick start, spec v1.1, benchmark details, and limitations documentation are provided.

Highlighted Details

• Benchmark Validated: AIGX is the only context format rigorously benchmarked, outperforming others on consistency and robustness across different AI models (Claude Haiku 4.5, Sonnet 4.6).
• Per-File Addressability: A dedicated index (files.aigx) provides targeted rule lookups per file, a key advantage over global or in-source methods.
• Zero Source Injection: Rules are stored externally in .aigx/, ensuring source code remains untouched.

Maintenance & Community

The project maintains spec v1.1 and publishes reference implementations across npm, PyPI, and crates.io. Users are encouraged to engage via GitHub issues and starring the repository.

Licensing & Compatibility

Tools are MIT-licensed, and the specification text is CC-BY-4.0. This permissive licensing allows for free adoption, reimplementation, and commercial use.

Limitations & Caveats

While AIGX excels in consistency and robustness, top formats can be a "statistical tie" in some metrics. AIGX serves as a "genome layer" that can be exported to simpler formats. Further details are in docs/limitations.md.