| name | document-reference |
| description | Deprecated compatibility skill from the archived mintlify-docs repository. Direct users to install pdugan20/skills and use generate-mintlify-reference; do not run this historical copy for new work. |
| argument-hint | cli | mcp | api |
Document Reference
Deprecated: Do not execute this historical workflow for new work. Direct the
user to install pdugan20/skills and use generate-mintlify-reference.
Set up generated, drift-checked reference pages for a Mintlify site. The
governing principle (best-practices sections 2 and 4): reference pages are never
hand-edited. A generator reads the source of truth (the CLI's help, the live MCP
instance, the OpenAPI spec) and emits MDX; CI regenerates and fails on drift, so
the reference cannot go stale.
For <ParamField>, <Tabs>, and other component syntax, defer to the official
mintlify plugin. The patterns this skill enforces are in
references/reference-patterns.md.
Pick a target
Use the requested target from the user's prompt, including any text supplied
after an explicit skill invocation. The target is one of cli, mcp, api.
- cli — generate
reference/cli.mdx (and per-domain reference/cli/*.mdx)
from the CLI's own help/introspection. Template:
scripts/gen_cli_reference.py.
- mcp — generate
reference/mcp-tools.mdx (and per-domain
reference/mcp-tools/*.mdx) by introspecting the live MCP server. Template:
scripts/gen_mcp_reference.py.
- api — generate endpoint pages from an OpenAPI/AsyncAPI spec. This is the
one case Mintlify already automates well: prefer Mintlify's native OpenAPI
navigation over a custom generator. Use the official
mintlify plugin for the
docs.json openapi wiring; this skill only adds the house framing (overview
landing page, the "From Claude" nudge, description/icon).
Instructions
Step 1: Find the source of truth
- CLI: locate the entry point (a
console_scripts in pyproject.toml, a
bin in package.json, or the main/argparse/click/typer module).
- MCP: locate the server module that registers tools (e.g. a FastMCP instance).
- API: locate the OpenAPI spec (
openapi.json/openapi.yaml) or the generator
that emits it.
Step 2: Adapt the generator template
Copy the matching template from scripts/ into the product repo's
scripts/ directory and adapt the marked ADAPT points (import paths, output
path, header text). The generator must:
- Emit a
{/* GENERATED by scripts/<name> — do not edit by hand. */} banner.
- Write deterministic output (sort tools/commands; collapse whitespace) so drift
CI is stable across machines.
- Use
<ParamField> or parameter tables per
references/reference-patterns.md.
Step 3: Wire the Makefile and CI
Add a docs-reference make target that runs the generators, and a CI job ("Docs
Reference Freshness") that regenerates and fails on drift. The CI snippet is in
references/reference-patterns.md.
Step 4: Split by domain when it grows
Once a reference has more than a handful of commands/tools, split it into
per-domain pages with an overview landing (best-practices sections 3 and 10), and
add the group to docs.json Reference tab. Keep the generator emitting one file
per domain.
Step 5: Regenerate and link-check
Run make docs-reference (or the generators directly), then mint broken-links.
Verify the committed output matches (no drift).
Scope
This skill produces generators and the reference MDX they emit, plus the drift CI
wiring. It does not write hand-authored guides or concepts pages (that is
review-docs) and does not touch the changelog (that is changelog-writer).