kud/claude-plugins/bookmark-search

Raindrop's search understands meaning, not just keywords — it matches across title, note, URL, tags and the page text itself. This skill uses that rather than keyword-matching, and narrows with filters instead of asking follow-up questions.

Step 1 — Resolve the query

If the user provided a query (e.g. /bookmark-search react hooks), use it directly.

If no query was given, ask: "What are you looking for?"

Step 2 — Search

Call find_bookmarks. Pass the user's own words as search — it is a natural-language parameter, so don't reduce a phrase to keywords. "that article about why CSS grid beat flexbox" works better whole than as css grid flexbox.

Translate anything else the user said into filters rather than a second search:

Leave sort alone unless the user wants oldest-first (created_asc) or a random sample (random) — the default already sorts by relevance for a semantic search.

Step 3 — Present results

Show the top 10: title, URL, collection, and tags where there are any. Link the title rather than printing a bare URL.

Semantic results are candidates, not exact matches — if the top hits look loosely related, say so rather than presenting them as certain.

If nothing comes back, don't just report zero. Say which filters were applied and offer the specific loosening that would help — dropping the collection filter, widening the date range, or searching without the tag.