| name | research-query |
| description | Searches across existing research briefs for claims, sources, or contradictions matching a query. Read-only — returns matching claims with cites, does not regenerate or re-fetch. Use to check whether a topic has already been researched, to pull specific facts from a prior brief, or to cross-reference claims across multiple briefs. Auto-triggers on: "/research-query", "do we have research on X", "what did the brief on Y say about Z", "find claims about", "search the briefs for", "have we already looked into". Do not use to run new research — use /deep-research for that.
|
| allowed-tools | Read, Grep, Glob |
Research Query
Search existing briefs without regenerating.
Inputs
<query> — free-text question or keyword- Optional:
--slug <slug> — limit to one brief
- Optional:
--min-confidence high|medium|low — filter by confidence tier
Execution
Step 1: Scope
If --slug given → read only state/briefs/<slug>/claims.json.
Else → glob state/briefs/*/claims.json.
Step 2: Match
For each brief, scan claims. Match <query> against:
claim text — direct substring and semantic overlapsub_question label (via trace.json if available)- Source URLs in
supporting
Rank matches: direct claim-text > source URL > sub-question.
Step 3: Return
Group by brief slug. For each match:
[<slug>] (freshness: <date>, verdict: <X>)
C<id>: <claim text> [S1, S3] (confidence: <tier>)
sources: <url>, <url>
If no matches → no matching claims across <N> briefs searched.
Cap total output at 600 words; if over, return the top 10 matches and append (<K> more matches hidden — run with --slug to narrow).
Rules
- Do NOT fetch new sources.
- Do NOT re-verify.
- Do NOT modify any brief.
- Stale briefs (
freshness > 30 days) are still returned but tagged (stale) — caller decides whether to re-run /deep-research.
- If a brief's
verdict is FAIL, tag (unverified) and do not surface its claims as authoritative.
Failure modes
| Code | Signature | Counter |
|---|
| F02 | Returned a "match" that isn't in any claims.json | Every returned line cites a real claim id from a real brief |
| F13 | Returned adjacent-but-irrelevant claims | Rank by direct match; drop below a relevance floor |
