What problem does it solve? AI-assisted coding produces fast diffs that are hard to review for hidden risks: behavior regressions, protocol field changes, architecture constraint violations, and hallucination-driven errors. This Skill turns a raw git diff into a structured risk review and, only after user confirmation, a classified commit message that PR reviewers can trust. ## Core Features & Use Cases - Seven-question risk review: Covers behavior changes, affected existing features, protocol/field changes, golden-test verification adequacy, AI uncertainty and hallucination risk types, change diffusion path, and architecture constraint breaches. - Four-tier risk classification: Classifies changes as Tuning, Behavior change, Protocol change, or Architecture change with review levels from Low to High, adopting the highest level when categories overlap. - Gated commit workflow: Never commits on the first pass; only commits after explicit user confirmation, then writes a commit message with classification, impact scope, verification evidence, and residual uncertainty. - Use Case: Before committing an AI-generated change to a planner routing module, run the review to surface that a session protocol field changed, note missing golden-test coverage, and produce a Medium-high classified commit message. ## Quick Start Ask the agent to review the current git changes for risk classification and then, after you confirm, write a classified commit message.
