What problem does it solve? AI-generated code often gets reviewed by the same session that wrote it, which rationalizes its own choices instead of investigating them. This Skill runs step 3 of the quorum pipeline: it reviews the built change from several independent lenses in fresh context, producing a panel of evidence for adjudication rather than a single self-agreeing opinion. ## Core Features & Use Cases - Six independent review lenses: behavior (operates the running app), correctness, spec-fidelity against plan.md, security, simplicity, and test-quality, each run in fresh context via separate subagents where available. - Structured review files: each lens writes a numbered findings file to docs/work/<slug>/reviews/ with severity levels (blocker, major, minor, nit), file:line references, and concrete failure scenarios. - Pipeline state tracking: records the reviewed commit head, round number, findings count, and any dropped lenses via the quorum state script so later steps know exactly what was examined. - Use Case: After a plan has been built, run the review step to get six blind reviews of the diff — including one that launches and drives the actual application — before adjudicating findings and opening a pull request. ## Quick Start Run the review step on a built work item by invoking /quorum:3-review with the work item's slug.
