| name | inference-brief |
| description | Render the top-of-context briefing for a target plugin. Reads the inference-engine catalog, filters elevated patterns tagged for the target plugin, writes state/briefings/<plugin>.md. Use before a session where the target plugin is about to do high-stakes work (e.g. /converge, /lich-review). Safe and cheap — rendering is a pure function of the current catalog. Auto-triggers on: "/inference-brief", "render the wixie briefing", "refresh briefings", "prep the ufopedia brief for <plugin>".
|
| allowed-tools | Bash(python *) Read Agent |
Inference Brief
Emit state/briefings/<plugin>.md — a concise Markdown summary of elevated patterns that apply to the target plugin.
Usage
The caller provides the plugin name. Defaults to wixie at Phase 1. Pass all to include every elevated pattern regardless of tag.
Pipeline
Step 1: Spawn the briefer agent
Delegate to the Haiku-tier briefer for a shape-check pass:
Agent(subagent_type="general-purpose", model="haiku",
prompt="Run the briefer agent defined at
wixie/plugins/inference-engine/agents/briefer.md
with plugin='<target>'.")
Step 2: Parse the agent's report
The agent returns one line:
rendered state/briefings/<plugin>.md (<N> elevated pattern(s), <M> bytes)
Step 3: Report to caller
Briefing rendered: state/briefings/<plugin>.md
Elevated patterns: <N>
Last reconciled: <timestamp from catalog>
If N == 0, the briefing file contains a placeholder explaining that no cross-session patterns have elevated yet — the caller should not block on this. The briefing is advisory.
Rules
- Do NOT render a briefing from a stale catalog. If the caller is about to do high-stakes work, run
/inference-reconcile first. If unclear, ask.
- Do NOT fabricate elevated patterns. If the catalog is empty, the briefing says so.
- Do NOT filter or sort differently from the renderer. Consistency across plugins is load-bearing.
- One briefing per file per plugin.
briefings/wixie.md is the only briefing a Wixie skill reads.
