A worked example of what /gather-agent-docs produces in an end-user project. The path layout matches what the skill writes into your own repo (.agents/docs/).
A small reference set for a fictional project that uses Gemini CLI and Cloud Run. The shape demonstrates the conventions:
INDEX.md— one dated entry per topic, plus a brief "How to use" section. The[YYYY-MM-DD]prefix is the oldestlast_verifiedamong that topic's files, so the index honestly reflects the staleness floor.gemini-cli/DOC.md— entry point in the default shape: frontmatter, one-paragraph summary, key concepts, a structured Documentation map (link index of upstream pages, grouped by upstream nav), and a "Subtopics" section. No captured snippets.gemini-cli/HOOKS.md— a subtopic split demonstrating the deepened shape: link map plus a small Patterns section with paste-ready snippets, each tied to an inline_Source: …_. Justified because hooks have their own canonical upstream page, get reached for repeatedly, and the deepened content would pushDOC.mdpast the soft cap.cloud-run/DOC.md— a topic that does not warrant a subtopic split: the documentation map fits comfortably within the soft cap, no patterns captured by default.
These pages exist to demonstrate the layout, not to ship truth. The map structure and URLs are correct in spirit but not actively re-verified on a schedule — copying them into a real project without re-running /gather-agent-docs (or /refresh-agent-docs afterwards) would mislead future agents about the staleness floor. Use the structure, regenerate the content.