Use AI assistants

Most code today is written with an AI assistant in the editor. Marreta is a new, focused language, so a model that has never seen it tends to guess, falling back to Python or JavaScript shapes that do not run. Marreta closes that gap by putting correct, authoritative context where assistants already look, generated from this documentation and the language’s own tested examples so it never drifts from the runtime.

In your project: AGENTS.md

marreta init writes an AGENTS.md at the project root. AGENTS.md is the cross-tool convention that agentic assistants load as in-context instructions. Marreta’s primer leads with the corrections a model most often needs (the things it gets wrong coming from another language), then a compact syntax cheat, then a pointer to the full reference.

Alongside it, init writes a thin pointer for GitHub Copilot, which reads its own .github/copilot-instructions.md rather than AGENTS.md. It points back to AGENTS.md, so there is a single source of truth.

To scaffold without the agent guide, pass --no-agents:

marreta init shop --no-agents

Refresh after upgrading

The primer is stamped with the runtime version it was generated for. After upgrading the runtime, regenerate it for the current project:

marreta agents

That rewrites AGENTS.md and the pointers from the installed runtime. marreta doctor reports when the file is behind, without ever changing it:

Agent guide:
  SKIP  AGENTS.md is stamped for 0.1.0, runtime is 0.2.0. Run `marreta agents` to refresh

On the web: llms.txt

Some assistants fetch documentation at query time instead of reading a project file. For them, the site publishes two files at its root:

• marreta.dev/llms.txt is a curated index of the documentation, one line per page with a short description.
• marreta.dev/llms-full.txt is the full reference, every guide page concatenated so a model can read it in one fetch.

Both are generated from this same documentation, so they stay current as the language grows.