Thought Wiki Corpus Fetcher

An agent-facing retrieval wrapper that queries a user's personal Thought Wiki (TW) corpus. It accesses three namespaces: Mind (the idea graph), Self (personality / decision journal), and Knowledge (the corpus that has been read and researched). It works through a single standard interface, via the tw CLI or the localhost HTTP API.

Note: The example output above is entirely fictional/representative; it does not contain any real person's notes. When you use the skill with your own corpus, the output will contain personal data — do not share real outputs publicly.

On the search side it applies hybrid retrieval — BM25 (sparse) + MiniLM embedding (dense) scores are combined with Reciprocal Rank Fusion, and each result returns with its source node id (cited). Nine modes: status, ask, search, persona, context, index, node, plan, schema.

Output: A row-based dataset consisting, for each result, of node_id, namespace, title, snippet, BM25 / vector / RRF scores, rank, and tags. If TW is down or the token is invalid, the skill does not produce a fabricated answer — it returns ok:false (hallucination prevention). It runs read-only; it makes no writes to the corpus.