AI and developer tooling

Start with the repository context

AGENTS.md points to docs/ai/PROJECT_CONTEXT.md and the verified architecture, design-system, component and development guides. These are snapshots: re-read the implementation relevant to the task before editing.

The repository includes specialized .agents/skills for component work, accessibility, theme integrity, documentation, testing and release preparation. Use the smallest relevant set, preserve local work and never claim tests ran when they did not.

A machine-readable documentation map

This site exposes /llms.txt with canonical guide, component and blog links. It is generated from the same content sources as the site, and describes the published packages and supported integrations. Use it to find the relevant reference, then inspect the actual public API and examples.

Make changes reviewable

Ask an assistant to identify the real component API, CSS source, tests and consumers before proposing a change. Regenerate normal build outputs, validate the affected app and inspect interactive states in a real browser. A passing typecheck does not prove visual quality.

Never provide private environment files or credentials as project context. Commit, push, package publication and CI changes require an explicit maintainer decision.

MCP is a future integration

There is no distributed Pimolu MCP server or public component-install CLI in the current repository. Do not invent an npx command, server URL or agent integration. A future server could expose verified APIs and examples; it should follow the same source-of-truth pipeline as this documentation.