docs-scraping

What I do

• •Turn a user docs scraping request into a reusable script and generated docs snapshot output.
• •Standardize output under docs/external/<source>/ with one page-level .ext.md file per discovered page.
• •Ensure each generated page has a metadata header and notes footer, and create an index.ext.md manifest.
• •Prefer routing implementation through the dedicated scraper subagent.

Accepted request patterns

Use this skill when the request includes one or more of:

• •A docs root URL (for example https://example.dev/docs)
• •A source/domain name (for example moonrepo, astro, pnpm)
• •A request to "scrape", "snapshot", "mirror", or "refresh" external docs

Output contract

• •Script: scripts/scrapes/scrape_<source>_docs.sh.ts
• •Output directory: docs/external/<source>/
• •Output files:
  • •Per-page files: <stable-page-stem>.ext.md
  • •Index file: index.ext.md

Behavior requirements

1. •Resolve source metadata from the user request:
   • •source key (safe folder/script slug)
   • •docs root URL
   • •discovery method (sitemap.xml preferred)
2. •Reuse project script conventions:
   • •Bun shebang (#!/usr/bin/env bun)
   • •*.sh.ts naming
   • •helpers under scripts/helpers/
3. •Use resilient scraping strategy:
   • •Primary: r.jina.ai markdown proxy
   • •Fallback: direct HTML fetch + conversion to markdown
4. •Normalize filenames from docs paths:
   • •deterministic flattening (for example docs__guides__intro.ext.md)
5. •Regenerate output cleanly:
   • •remove old *.ext.md in target source directory
   • •write fresh per-page files and index.ext.md

Generated page format

• •Top ---- section with summary metadata:
  • •captured timestamp
  • •source root
  • •source page URL/path
  • •keywords
  • •concise summary
• •Middle body with markdown content snapshot
• •Bottom ---- notes/comments/lessons section

Index format

• •Top ---- section with source/capture metadata and summary totals
• •Full page inventory with links to local page files and capture status
• •Bottom ---- notes/comments/lessons section

Implementation workflow

1. •Inspect existing scraper scripts for reuse patterns (scripts/scrapes/scrape_*.sh.ts).
2. •Create or update scripts/scrapes/scrape_<source>_docs.sh.ts.
3. •Run the script once to generate docs output.
4. •Report totals (pages, ok, failed) and notable blocked pages.
5. •If a new script entrypoint was introduced, update README.md and AGENTS.md.

Safety constraints

• •Never embed secrets, auth headers, or private tokens in script or output files.
• •Skip private/authenticated docs pages unless explicit credentials handling is requested and safe.
• •Keep scripts idempotent and deterministic where practical.

Return checklist

• •Script path(s) added/updated
• •Output directory generated/refreshed
• •Discovery strategy used
• •Capture totals and notable failures
• •Exact rerun command (bun scripts/scrapes/scrape_<source>_docs.sh.ts)