What problem does it solve? Code review output is often noise: unranked lists of things that look wrong, with no proof they are real defects. This Skill produces a small number of verified findings, each with a required behavior, a feasible trigger, a concrete contradiction, an observable consequence, and the strongest counterargument already checked. ## Core Features & Use Cases - Agreement-based correctness analysis: Anchors review on agreements between participants across boundaries (caller/callee, producer/consumer, writer/reader) rather than scanning file by file, surfacing the cross-boundary defects reviewers miss most. - Three selectable dimensions: Correctness is the default core; performance (workload to resource demand to consequence) and security (source to trust boundary to sink) activate as sub-cases of the same engine via the dimensions argument. - Refutation discipline: Every candidate finding must survive a five-part test — supported obligation, feasible execution, concrete contradiction, observable consequence, and an examined counterargument — before it is reported. - Use Case: After a large refactor touching async task handling, run the review on working changes to force violating executions around identity correlation and authority reconciliation, then receive ranked findings citing both sides of each broken agreement. ## Quick Start Ask the AI to review the current working changes for correctness defects using the code-review skill, optionally adding performance or security dimensions.
