AI agents

If you build with Claude Code, Cursor, Copilot, or any other coding agent, give the agent the same conventions this documentation teaches you. Alacris ships a ready-made instructions file for that:

⬇ Download AGENTS.md

Put it in your project root as AGENTS.md — the emerging convention that Claude Code, Cursor, Codex and others read automatically — or paste its contents into CLAUDE.md, .cursorrules, or whatever your tool uses.

curl -o AGENTS.md https://bmartel.github.io/alacris/AGENTS.md

What it covers

The file is self-contained — an agent needs no other context to produce idiomatic Alacris code:

• The mental model, stated as hard rules: a function in ${} is a live binding, setup runs once, there is no re-render.
• Both setup paths — a no-build import map and the npm/bundler route — and the instruction not to introduce a build step into a project that doesn’t have one.
• Project organization: one component per file, tag-name prefixes, where stores, contexts and shared style tokens live.
• The full binding syntax, each and keying, store / selector, context, and the styling and theming contract (vars, ::part, never !important).
• Security rules the agent must not break — no untrusted .innerHTML, scheme-validate URLs, keep runtime values out of css templates. See Security for the model behind them.
• A wrong → right table of the mistakes agents (and people) actually make, and a verification checklist to run before declaring a task done.

For search and retrieval

The site also serves llms.txt — a machine-readable map of this documentation following the llms.txt convention — so agents that fetch documentation on demand can find the right page in one hop.