llms.txt

# AIGX - AI Genome Exchange

> AIGX (AI Genome Exchange) is an open, MIT-licensed context format for AI coding agents. It stores a
> codebase's rules in a centralized `.aigx/` directory and adds a per-file "boundary index" - for each
> source file, the exact rules, forbidden imports, and gotchas that apply there. It is the only context
> format known to be validated by a controlled benchmark: #1 of all formats tested on mean score, pass@1,
> and hidden-test pass rate on both Claude Haiku 4.5 and Claude Sonnet 4.6 at n=60, having survived ~24
> deliberate attempts to beat it. The metaphor: a genome is the central, portable context that builds and
> operates an organism - AIGX is the genome of your codebase, which any AI agent reads to inherit your
> conventions, while your source code stays untouched.

## Core docs

- [README](README.md): overview, proof, quick start, comparison, FAQ.
- [Specification (v1.1)](SPEC.md): the normative format definition - directory layout, file grammar, rule
  ids, the per-file boundary index, the agent addendum, conformance, comments clause, monorepo scaling.
- [Benchmark](BENCHMARK.md): full methodology, n=60 results on two models, the challenger log, and the
  honest caveats.
- [Changelog](CHANGELOG.md): spec version history with migration notes.

## Concepts and guides

- [Concept - why "genome"](docs/concept.md): the biological metaphor and why it predicted the winning design.
- [Principles](docs/principles.md): seven benchmark-backed laws for writing context for AI agents.
- [Authoring guide](docs/authoring-guide.md): how to write a genome for your codebase, with a checklist.
- [FAQ](docs/faq.md): common questions, comparisons to AGENTS.md / CLAUDE.md / Cursor rules / llms.txt.
- [Limitations](docs/limitations.md): honest scope, threats to validity, and responses to critiques.
- [Migration guide](docs/migration.md): how to adopt AIGX alongside an existing AGENTS.md / CLAUDE.md.
- [Glossary](docs/glossary.md): terms and definitions used in the spec and docs.

## Example and templates

- [Worked example](examples/sourcing-app/): a complete real-world genome (.aigx/ + per-domain cards).
- [Minimal example](examples/minimal/): the smallest valid genome - 3 files, one rule, one index entry.
- [Starter templates](templates/starter/.aigx/): copy into your repo to begin.

## Tooling

- [aigx-lint](tools/aigx-lint/): zero-dependency Python validator + resolver. CI-ready. `--resolve PATH`
  returns one file's boundary entry in O(1).

## Key facts

- Spec version: v1.1 (stable, normative).
- License: MIT (free to use, fork, and build on).
- Format: plain-text XML-style tags in a `.aigx/` directory; nothing injected into source code.
- Integration: one instruction line added to any agent's config (AGENTS.md, CLAUDE.md, Cursor rules, etc.).
- Differentiators: per-file boundary index, zero source-code injection, tool-agnostic, benchmark-validated.
- Honest caveat: the top context formats are statistically tied on the mean; AIGX's edge is cross-model
  consistency, robustness under challenge, simplicity, and being the only format that was measured at all.