docs-detect-source — Identify and route a docs source

Workflow#

1. Inspect the input: classify as a plain URL, a GitHub repo URL, or a local path.
2. For a plain URL — fetch the page and look for platform signals in HTML meta tags, CDN links, and domain name. If no platform matches, classify as a generic website.
3. For a GitHub repo URL — fetch root contents via the GitHub API and look for known config files (mint.json, SUMMARY.md, docusaurus.config.js, .vitepress/, astro.config.*, etc.). If only a README and source directory are present, classify as a code repo.
4. For a local path — inspect the directory structure for the same config file signals.
5. Map the detected type to a downstream skill and report both.

Guardrails#

• Fetch the page with WebFetch first; only escalate to a browser if the HTML is empty.
• A single ambiguous signal is not enough — require at least one platform-specific config file or meta tag before assigning a platform type.
• If detection is inconclusive, default to website / /docs-from-site rather than blocking.
• Never mutate the source; detection is read-only.

Skill Routing#

Acceptance Criteria#

• Input classified into exactly one type without prompting the user
• Platform identified from signals (meta tags, config files, domain) rather than guessing
• Downstream skill name reported alongside the detected type
• Inconclusive inputs default to website gracefully