| name | inference-query |
| description | Search the inference-engine catalog. Accepts a code (F07, OP05, …), a tag (wixie, lifecycle, …), or a 16-char pattern_id. Returns the full pattern records as JSON. Read-only; cheap; safe to call ad-hoc. Auto-triggers on: "/inference-query", "what does F07 say", "show patterns tagged lifecycle", "search ufopedia for …".
|
| allowed-tools | Bash(python *) Read |
Inference Query
Retrieve pattern records from catalog.json by code, tag, or pattern_id.
Usage
The caller provides one search term. Exact match only at Phase 1. Fuzzy / BM25 ranking is deferred.
Pipeline
Step 1: Run the query
python ${CLAUDE_PLUGIN_ROOT}/../../shared/scripts/inference-engine.py query <term>
The engine returns a JSON array of matching patterns to stdout. Exit code 0 if any match, 1 if none.
Step 2: Summarize for the caller
Parse the JSON and present a compact summary:
<N> pattern(s) matching '<term>':
[elevated] F07 — Ran seven reactive iterations instead of one-pass Wixie lifecycle
weight 0.94 · posterior 0.83 (CI 0.55–0.98) · LLR 8.95
last seen 2026-04-21 (1d ago) · 5 observations across 1 session
tags: wixie, lifecycle, convergence, prompt-engineering, process-discipline
Step 3: Offer follow-ups
If the caller's intent is to understand a pattern they're about to trip over, offer:
/inference-brief <plugin> to see the full briefing the target plugin is consuming.- Full JSON via a Read on
wixie/plugins/inference-engine/state/catalog.json.
Rules
- Do NOT guess at match results. If the engine returns empty, tell the caller honestly.
- Do NOT paraphrase the pattern's
signal or counter. They were written to be re-read verbatim.
- Do NOT expose non-elevated patterns as if they were elevated. The verdict field is honest; keep it visible.
