What problem does it solve? When the same AI session that made a decision also reviews it, the review inherits the author's blind spots and defends its own reasoning. This Skill removes that bias by delegating the review to a separate subagent with a clean context and a different model, so artifacts claiming to be "done" get a genuinely independent check. ## Core Features & Use Cases - Clean-context review: The reviewer subagent receives only file paths, never a summary of why decisions were made, eliminating confirmation bias. - Different-model verification: The subagent runs on a model different from the current session, since blind spots are embedded in model weights, not just context. - Structured findings: Findings use severity labels (critical / discuss / non-blocking), a maturity score given before conclusions, and a JSON verdict of approved, rework, or blocked. - Use Case: A product document or code slice has grown large and its owner doubts it. The Skill identifies the review object (via git diff against the task_spec baseline), assembles relevant context files from the artifact registry, launches the subagent, and delivers the verdict unfiltered to the owner. ## Quick Start Ask the assistant to run an independent review of the document or code you just finished before marking it as done.
