What problem does it solve? A single reviewer or single-pass critique often misses failure modes because one perspective anchors the analysis. This Skill challenges a plan, decision, design, or code diff with several independent lenses that never see each other's reasoning, then merges their findings into one deduplicated, severity-ranked verdict. ## Core Features & Use Cases - Multi-lens fan-out: Runs independent read-only lenses (red-teamer, code-reviewer, independent-revalidator, brainstormer) chosen per artifact type from a config file, each returning a normalized JSON finding contract with anchor, severity, and fix. - Consolidated verdict: A consolidator dedups cross-lens findings, ranks by severity, attaches repeat-offense metadata, and proposes a PASS, PASS_WITH_RISK, or BLOCKED verdict written to a markdown report. - Gate and loop modes: Optional --gate writes a machine-readable critique-consensus.json verdict artifact for downstream stage gating, and --loop runs a bounded critique-refute cycle for contested verdicts. - Use Case: Before committing to an architecture decision, run the critique on the design document so adversarial, validation, and product lenses independently attack it, then review one ranked report of blockers and DEC-worthy items. ## Quick Start Ask the AI to critique the plan at plans/my-design.md using the critique skill and report the consolidated verdict.
