reward-guard

Task completion requires evidence-based validation through three mandatory checks: technical validation, consistency verification, and completeness assessment. This Skill enforces all three before allowing state transitions to done, blocking premature declarations.

Scope creep detection identifies unauthorized changes to task boundaries during completion. The Skill verifies that task scope remains aligned with original specifications, preventing hidden expansions that invalidate completion criteria.

Critic agent review provides independent validation that evidence meets completion standards and state transitions are justified. This actor-free control ensures objective assessment before marking tasks done, eliminating self-reward bias.

Yes. The Skill activates via pre-tool scripts for trigger-based invocation, automatically enforcing validation and critic review whenever task completion is attempted. This creates an auditable record of all state transitions.

Failed validation blocks the state transition to done. The Skill returns specific validation failures—technical, consistency, or completeness gaps—allowing you to address deficiencies before resubmitting for completion.

The Skill depends on jq for data processing and requires a critic SubAgent for review. If you have jq installed and access to a critic agent, the Skill integrates directly into your completion workflow.