What problem does it solve?
It prevents premature acceptance of a strategy, plan, spec, or completion claim by identifying missing evidence, hidden assumptions, contradictions, and verification gaps before they cause downstream rework or risky decisions.
Core Features & Use Cases
- Confidence guardrail for engineering artifacts: Reframes “are we confident enough?” into a structured stress-test with explicit confidence levels and required evidence.
- Material loophole detection + repair actions: Surfaces only loopholes that can change scope, design, acceptance criteria, security/privacy/reliability, or rollout decisions, and proposes repairs (edit artifacts, inspect repo, fetch docs, ask the user, or mark as Open Question/Unverified).
- Loop-limited reconciliation workflow: Runs 1 loop by default with a hard limit of 3 loops, stopping early when remaining items are properly classified as Open Question/Unverified.
Use cases: validating an interview packet; between spec and plan; before freezing implementation/task breakdown; after subagents return but before reconciliation acceptance; before claiming “complete/fixed/verified/ready/accepted” after code review, security audit, or test reports.
Quick Start
Use the strategy-stress-test skill on your current strategy/spec/plan (or completion claim) and ask for a confidence verdict with listed material loopholes and concrete repair actions.