Build with an AI agent

These docs are published in a machine-readable form, so an AI coding agent can read the whole SDK and build a plugin for you. Install the package, point your agent at one URL, and describe what you want.

Machine-readable docs

File	What it is
/llms.txt	A short index with links — the llmstxt.org standard.
/llms-full.txt	The entire documentation as one clean Markdown file.
/llms-small.txt	An abridged version that fits smaller context windows.

The workflow

1. Install the SDK:

   pip install mixlar-sdk

2. Give your agent the docs. Paste something like this to Claude, Cursor, Codex, or Copilot:

   Read https://docs.mixlar.net/llms-full.txt — it’s the full Mixlar plugin SDK reference. Then scaffold a plugin with mixlar-sdk create, build the idea I describe below, validate it with mixlar-sdk validate, and render the widget with mixlar-sdk emulate --render preview.png. My idea: a plugin that shows my unread email count on a widget and has a macro to mark all read.

3. Let it build, then test with the emulator — no hardware needed:

   mixlar-sdk validate
   mixlar-sdk emulate --render preview.png

Tip

Tools that natively understand llms.txt (Cursor, and some Claude / Copilot setups) can just be pointed at docs.mixlar.net — they discover /llms.txt and pull only the pages they need.

Note

Prefer a self-contained context file? Every page here is also in llms-full.txt, so you can download it once and hand it to an offline model. The mixlar-sdk CLI itself is the other half — the agent runs create / validate / emulate / pack to actually build and test.