Read the documentation as Markdown
Every documentation page has a plain Markdown version, and the site publishes
an llms.txt file that maps the documentation. They exist for coding agents
and scripts that already use BenchBox and need to read a page without parsing
HTML. They do not help an agent find BenchBox in the first place.
The llms.txt file
https://benchbox.dev/llms.txt follows the llms.txt convention.
It opens with a one-paragraph summary, then lists links in sections:
- Install and first run.
- CLI reference.
- MCP reference.
- Methodology.
- Result summaries.
- Every page that
docs/nav.ymllists, grouped like the sidebar. - Optional: pages that the navigation does not list.
The build generates the file from docs/nav.yml and the converted pages. A page
that the navigation omits appears under Optional, so a new page cannot go
missing. Generated per-query and compatibility pages are not listed one by one;
the listed index pages link to them.
The Markdown version of a page
Replace the .html suffix of a page URL with .md:
| Page | Markdown version |
|---|---|
https://benchbox.dev/docs/usage/installation.html | https://benchbox.dev/docs/usage/installation.md |
https://benchbox.dev/docs/tutorials/index.html | https://benchbox.dev/docs/tutorials/index.md |
The Markdown file starts with the page title. Links to other documentation
pages point at their Markdown versions with the same #fragment. Links to
anything else are absolute URLs. Callouts keep the GitHub alert form
(> [!NOTE]), diagrams stay in mermaid code fences, and the invisible
anchors that the HTML pages use are removed.
GitHub Pages serves .md files as text/markdown; charset=utf-8 and .txt
files as text/plain; charset=utf-8.
The Markdown versions cover the documentation pages under /docs/. The blog,
the result summaries under /result-summaries/, the search page and the
Results Explorer are HTML only. An interactive results embed becomes a link to
the HTML page.