Handbook
forge-evidence-review
Use when outputs need to become human-reviewable evidence.
Procedure: 1. Gather AgentRun, LCDLTraceSummary, FleetJobSummary, and artifact refs. 2. Classify status as pass/warning/fail/needs_decision. 3. Identify risks and missing evidence. 4. Produce EvidencePacket-compatible output.
Output
Return a concise summary, risks, evidence refs, and the next human decision.