What problem does it solve? Code reviews done in the main conversation context lose objectivity and miss project-wide issues like duplication, architectural drift, and convention breaches. This Skill runs the review in a forked subagent with no conversation history, loads project conventions and task context first, and produces a persistent written report instead of ephemeral chat output. ## Core Features & Use Cases - Scoped Reviews: Review staged changes, all changed files, all Python files, a specific package, or an explicit file list, with an optional ticket ID to load task context. - Architectural Drift Analysis: For larger change sets, runs a separate pass detecting blast-radius imbalance, logical coupling via git co-change history, unassigned new files, boundary crossings, and test-cost trends. - Structured Report Output: Writes findings grouped by Critical / Major / Minor / Suggestions to a dated report file in the project's basic-memory directory, returning only a short summary to the main agent. - Use Case: After finishing a feature branch touching eight files, invoke the review with scope "changed TOO-14" to get correctness, security, and duplication findings plus an architectural drift assessment tied to the ticket's intent. ## Quick Start Ask the agent to review the staged changes with the code-review skill and write the findings report.
