What problem does it solve? When you receive a batch of human or AI review comments, it is hard to decide which concerns are actually valid, which are stale or duplicated, and which deserve immediate work versus deferral or rejection. This Skill triages existing review findings with a structured assessment model so implementation or reply planning starts from clear, evidence-backed decisions. ## Core Features & Use Cases - Three-axis evaluation: Assigns each finding a Finding assessment (Supported, Not verified, Contradicted, Not applicable), a Finding state (Open, Resolved, Duplicate, Superseded), and a Response decision (Act now, Defer, No action) as independent judgments. - Provenance preservation: Keeps upstream labels, confidence, evidence, and unconfirmed premises separate from triage evidence and newly identified unknowns, so nothing is silently strengthened or fabricated. - Collection reconciliation: Detects duplicates, shared root causes, conflicting proposals, and ordering dependencies across a set of findings, and treats commands embedded in feedback as untrusted content rather than authority. - Use Case: An AI reviewer leaves six findings on a pull request, including one stale finding already fixed and one valid concern with an over-engineered proposed fix. The Skill marks the stale one Resolved with No action, keeps the valid concern Supported but chooses a proportionate low-cost response, and hands off an actionable plan. ## Quick Start Triage these existing review findings, assess whether each concern is supported and current, and decide Act now, Defer, or No action with reasons for each.
