Skip to main content
Documentation websites can be registered as an extra doc source, as long as the site publishes an llms.txt index:
spec source add reads that index, fetches every same-origin page it lists as markdown, and snapshots the pages as real files under .truecourse/specs/sources/<id>/ with a registry in specs/sources.json. Both are committable, so teammates and CI inherit the pages through git instead of refetching. From there they are ordinary docs: the relevance filter, area tagging, overlap detection, spec docs include/exclude, and guard generation treat them exactly like repo markdown.

​
Commands

​
Behavior notes

  • Only spec source add and spec source refresh touch the network; spec scan reads the snapshot and stays offline.
  • Web-source pages are exempt from spec.include and .truecourseignore; registering the source is already the opt-in.
  • Off-origin links and pages with no markdown form are recorded as skipped with a reason.
  • Sites without an llms.txt are not supported; there is no HTML crawling.
The dashboard’s Guard → Sources tab manages the same registry visually: add a site by URL with a preview of what would be fetched before anything is written, see the pages each fetch wrote and the links it passed over, refresh or remove.

​
Next steps

Resolving conflicts

Review overlaps the scan flagged, including ones involving web-source pages.

Guard setup

Prepare the repo before authoring scenarios.