What problem does it solve? An agent that authors an integrations prior-art survey cannot be trusted to grade its own work, and a deterministic validator can only check structure — not whether a locator host really belongs to the vendor, whether an evidence quote supports its claim, or whether a classification was faithfully transcribed. This Skill is the human-judgement half of a two-part gate that reviews integration vocabulary maps, angle search outputs, extract records, and the integration register before they are accepted. ## Core Features & Use Cases - Thirty numbered review conditions: Judges vendor-host resolution, authority bands, verbatim query and quote fidelity, enumerated-versus-bounded zeros, admission tests, complexity component weights, lens denominators, and absence-claim phrasing — each finding tied to a named condition. - Thirteen-source evidence protocol: Reads the artifact alongside the producer package's schemas, source registry, angle references, absent-input policy, category vocabulary, extraction and synthesis guides, and the scope the producer was handed. - Single verdict output: Emits exactly one line — VERDICT: approve or VERDICT: revise — with every revise finding naming its condition, and explicitly refuses to false-revise honest small results. - Use Case: After running the integrations-prior-art-survey producer on a new product scope, run this reviewer on the resulting integration register to catch a fabricated classification or an inflated first-party authority band before the register drives build prioritization. ## Quick Start Use reviewing-integrations-prior-art-survey to review the integration register and report in my survey output directory and give me the verdict with findings.
