wiki-query

name	wiki-query
description	Query pro-workflow wikis via SQLite FTS5 BM25 retrieval. Returns top-K passages with citations. Use when answering a question that any of the user's wikis already covers, when the user says "what does the wiki say about X", "ask wiki", "search wikis", or before drafting a new wiki page (to avoid duplication).

Wiki Query

FTS5 BM25 retrieval over wiki pages indexed by wiki-builder.

When to use

• Before writing any new wiki page → check coverage first
• User asks a domain question that may already live in a wiki
• "Ask the wiki: "
• Verifying citations before quoting a claim
• SessionStart auto-load when prompt matches a known wiki topic

Commands

node $SKILL_ROOT/scripts/query.js search "<query>" [--wiki <slug>] [--limit 10] [--json]
node $SKILL_ROOT/scripts/query.js related <slug> <rel-path> [--limit 5]
node $SKILL_ROOT/scripts/query.js show <slug> <rel-path>

search with no --wiki ranks across all wikis. related finds adjacent pages by reusing the page's title + summary as the query.

Output

JSON-friendly. Each hit:

{
  "page_id": 12,
  "wiki_slug": "agent-memory",
  "rel_path": "wiki/concepts/episodic-memory.md",
  "title": "Episodic Memory",
  "snippet": "... [time-stamped] traces, distinct from semantic ...",
  "rank": -3.21
}

Lower (more negative) rank = better BM25 match.

Citing back

Every wiki hit must be cited as:

[wiki:<slug>] <title> — `<rel_path>`

Do not paraphrase a hit without showing the source.

SessionStart integration

When pro-workflow's SessionStart hook detects wiki-relevant terms in the user prompt, it runs query.js search "<prompt>" --limit 3 and injects top hits into the session as a hint:

[wiki-query] 3 relevant pages:
- agent-memory · wiki/concepts/episodic-memory.md
- agent-memory · wiki/papers/park-2023-generative-agents.md
- ...

Helps Claude recall existing knowledge instead of redoing research.

Limits (Phase 3.3.0)

• BM25 only. Vector search arrives 3.3.2 with sqlite-vec.
• No re-ranking. MMR diversity arrives with the research loop in 3.3.1.
• Snippet window is 16 tokens around match — tune via --snippet-len.