What problem does it solve? Code review agents often produce inconsistent severity judgments—either missing critical security issues or flooding reports with low-value style opinions. This Skill defines a three-tier classification contract that standardizes how strictly each review agent must report findings. ## Core Features & Use Cases - Three-Tier Agent Classification: Groups review agents into Blocking (security, data migrations, deployment), Analytical (architecture, performance, language-specific reviewers), and Advisory (simplicity, schema drift) tiers, each with distinct confidence thresholds and reporting rules. - Conclusion Type Overlay: Adds a semantic conclusion_type field (finding, question, needs-human-check, clear) mapped onto the upstream findings schema without modifying it. - Anti-Leniency Rules: Lowers the P0 reporting threshold to confidence 0.50 for Blocking agents, requires evidence with file paths and line numbers, and downgrades Advisory-tier style opinions to P3. - Patch Gate Integration: When ce:review runs with the [T] flag, the Deterministic Patch Gate automatically consumes tier classifications to force Blocking findings into gated_auto. - Use Case: Before running ce:review on a pull request touching database migrations, load this contract so the data-migrations-reviewer applies zero-tolerance Blocking rules while the code-simplicity-reviewer stays advisory. ## Quick Start Load the review-contract skill before running ce:review so the review agents apply the correct tier-based strictness to this pull request.
