verify-agent-claims

Ground-truth verification discipline for agent and subagent output. Triggers whenever integrating, acting on, or reporting a claim produced by an AI agent — a completion report, a state assessment, a justification for an action, or a 'this is safe to delete/merge/deploy' assertion. Defines claim classes, the verification move for each, and the escalation rule when verification fails. Distilled from the 2026-06 product-factory programme, where this discipline caught 10+ confident-but-wrong agent claims before they caused damage.