Summary
Add an llms.txt (and optionally llms-full.txt) at the docs site root so LLM agents have a curated, machine-parseable entry point to ConfUSIus's documentation.
Background
llms.txt is a proposed standard (Jeremy Howard / Answer.AI, Sep 2024) for giving LLMs a concise entry point to a site's docs. Adopted by FastAPI, Next.js, Vite, Vue, Cloudflare, Mintlify, and many others.
It is a single Markdown file at the site root with a fixed structure (H1, blockquote, optional prose, then H2-delimited link lists). LLMs use it to discover what's available; humans can read it too.
Proposal
- Hand-curate
docs/llms.txt in the repo, with sections like:
## User Guide — links to the main user-guide pages
## API Reference — links to the mkdocstrings-rendered API pages
## Examples — links to the Jupytext-built example gallery
## Changelog — link to the changelog
## Optional — links to external references (BIDS spec, xarray docs, etc.)
- Submit the deployed URL (
https://confusius.tools/llms.txt) to the llmstxt.site directory for discoverability (one-click).
- (Optional follow-up) Generate
llms-full.txt in CI by walking site/**/*.html, converting to markdown, and concatenating — gives agents the full ~45-page context in one fetch.
Why
- ~45 pages of docs (user guide, API reference, examples) are currently invisible to LLMs unless they browse
- A clean
llms.txt makes the package discoverable and usable by Claude Code, Codex, Gemini CLI, Goose, Cursor, Copilot, etc.
- Deploys automatically via the existing
.github/workflows/docs.yml deploy-docs job — no new infra
- Low effort, high signal
Effort
Low. The hand-curated file is ~30 min; the generator + CI hook is a follow-up.
Summary
Add an
llms.txt(and optionallyllms-full.txt) at the docs site root so LLM agents have a curated, machine-parseable entry point to ConfUSIus's documentation.Background
llms.txt is a proposed standard (Jeremy Howard / Answer.AI, Sep 2024) for giving LLMs a concise entry point to a site's docs. Adopted by FastAPI, Next.js, Vite, Vue, Cloudflare, Mintlify, and many others.
It is a single Markdown file at the site root with a fixed structure (H1, blockquote, optional prose, then H2-delimited link lists). LLMs use it to discover what's available; humans can read it too.
Proposal
docs/llms.txtin the repo, with sections like:## User Guide— links to the main user-guide pages## API Reference— links to the mkdocstrings-rendered API pages## Examples— links to the Jupytext-built example gallery## Changelog— link to the changelog## Optional— links to external references (BIDS spec, xarray docs, etc.)https://confusius.tools/llms.txt) to the llmstxt.site directory for discoverability (one-click).llms-full.txtin CI by walkingsite/**/*.html, converting to markdown, and concatenating — gives agents the full ~45-page context in one fetch.Why
llms.txtmakes the package discoverable and usable by Claude Code, Codex, Gemini CLI, Goose, Cursor, Copilot, etc..github/workflows/docs.ymldeploy-docsjob — no new infraEffort
Low. The hand-curated file is ~30 min; the generator + CI hook is a follow-up.