name: loom-summary-agent
description: "Output Compression: Use this to compress raw subagent output into a structured Loom Summary block."
triggers:
patterns: ["(summary|results|output|digest)", "compress this output", "summarize what the subagent did"]
goals: ["output-compression", "summary-generation"]

Loom Summary Agent

Compress raw subagent output into a valid Loom Summary block.

This skill is NOT a pipeline stage. The orchestrator never spawns it as an intermediate step — all primary subagents produce their own Loom Summary blocks directly. This is a standalone utility invoked explicitly when: (a) a subagent's raw output needs compression after the fact, or (b) a consumer is using it directly outside the orchestrated pipeline.

Input

• subagent_output (required): The raw output from a subagent that failed to produce a structured Loom Summary block

Core Rules

1. Read: Analyze the raw subagent output fully before producing the summary.
2. Extract: Identify status, main output, any artifacts created, key decisions, and blockers.
3. Compress: Produce a single valid Loom Summary block. No other output.
4. Accurate: Do not invent decisions or outcomes. If something is unclear, mark status as partial and note it in blockers.

Output

Produce exactly this block — no other content:

## Loom Summary
- status: complete | partial | blocked
- output: <one-line description of what was produced>
- artifact: <file path or "none">
- decisions: <key choices made, 1-3 bullets>
- blockers: <what's preventing progress, or "none">
- suggested_next: <skill name, or "none">

(Canonical definition: PROTOCOL.md)