Used to proceed with simple investigations in a refmesh-first manner. Standardizes the process of executing `refmesh-search` first, determining hit status, falling back to primary sources, deciding between `refmesh-register` / `refmesh-curate`, and escalating when the task cannot be closed by investigation alone.
Use whenever a `refmesh register` summary surfaces a `⚠ Similar existing concepts` warning, when search results show clearly outdated or duplicated knowledge, or when the user asks to clean up / forget / merge concepts. Covers the four lifecycle moves — SAME_AS merging, REPLACES/DEPRECATES tagging, archive (logical hide), and prune (physical delete).
Use when the user asks you to fetch / open / analyze / summarize a URL, documentation page, blog post, paper, RFC, or codebase fragment and remember / store / shelve / 蓄積 / 覚えて its concepts. Fetches the source if a URL is given, extracts concepts and typed relationships, links them to concepts that already exist in the local refmesh graph, and persists everything through `refmesh register` so later sessions can recall it through `refmesh search`.
Use at the start of any task that touches a topic the user has previously asked you to remember, or whenever you suspect prior knowledge exists in the local refmesh graph. Pulls a connected slice of concepts, relationships, and source references through `refmesh search` so you can answer with grounded context instead of guessing.