docs trained

mental model

Treat documentation as an interface with an owner. The publisher's current documentation proves an API claim; llms.txt is an index; a local reader is only a reading aid. Mintlify owns a Mintlify site's configuration and build.

examples

# Discover a library entry, then open the publisher's linked documentation.
node moe-docs/references/docs-seeker/scripts/detect-topic.js 'Next.js caching'

# Build an index, then inspect it and its links.
python3 moe-docs/references/llms/generate-llms-txt.py --source docs --output .

# For a Mintlify site, use only commands exposed by the installed CLI.
mint validate && mint broken-links && mint a11y

best practices

Cite a primary page next to every mutable API or CLI claim.
Generate an index from source and inspect its real URLs before publishing it.
Run mint validate after a Mintlify change; follow with link and accessibility checks when they cover the changed surface. Mintlify Quickstart

strengths

This guru separates research, indexes, documentation sites, and local review so each task uses the owner that can verify it.

weaknesses / pain points

llms.txt has no authority over its links, Context7 coverage is incomplete, and a local Markdown reader cannot prove a deployed documentation site.

gotchas

A Markdown URL ends before its closing ). The analyzer strips it before the link is fetched.
The current Mint CLI package is mint, not mintlify.
Use commands advertised by mint --help; do not carry old CLI names forward.

known bugs

No version-specific upstream defect is recorded here. The local analyzer and generator quirks that matter are listed under gotchas.

troubleshooting

symptom	root cause	fix
an API answer is stale	a discovery index was treated as authority	open and cite the publisher's current documentation
Mint command is unavailable	the command came from an outdated guide	run `mint --help`, then use the supported command
index link has a trailing `)`	Markdown punctuation was parsed as URL content	the analyzer strips Markdown punctuation from URLs, and its regression test covers the case
generator finds no docs in scratch	it inspected absolute path segments for hidden names	inspect paths relative to the requested source

practiced cases

In a local Node run, detect-topic.js 'How do I use cache in Next.js?' returned next.js and cache, and the three-case test suite passed. Fed a Markdown link, the analyzer strips the closing punctuation instead of returning https://example.com/guide):.
Mint CLI 4.2.940 advertises dev, validate, broken-links, and a11y through npx --yes mint --help. Commands it does not advertise stay out of the routing.
The Python generator handles a Markdown source in the scratch folder by inspecting paths relative to the requested source: the false "No markdown files" result does not appear, and it produced a one-entry llms.txt pointing at https://example.com/docs/guide.

sources

Read the docs guide.