What problem does it solve? When a worker agent or main session reports a task as complete, standard code review only validates the happy path and can silently share the same blind spots as the worker. This Skill forces an adversarial review pass that tries to disprove the completion claim before it is accepted. ## Core Features & Use Cases - Attack Vector Checklist: Provides 10 concrete attack vectors (missing artifacts, weakened test assertions, lost activity records, scope creep via git diff, Goodhart metric gaming, and more) to systematically challenge completion claims. - Independence Constraints: Documents workarounds for the delegate_task API limitation where sub-agents share the parent model, including framing changes, terminal-enabled toolsets, and separate hermes processes. - Decision Gating: Routes findings back to the worker as new requirements when an attack succeeds, or forwards verified completions to the correction-funnel knowledge base. - Use Case: After a worker reports "Status: success" on a multi-file refactor, trigger this Skill to run the attack vectors, verify artifacts actually exist on disk, and confirm test assertions were not weakened before accepting the handoff. ## Quick Start After the worker reports the task complete, run an adversarial review on its handoff and try to disprove the completion claim using the attack vector checklist.
