review-changes

Run change detection to get risk-scored changes, then query affected execution flows and impact radius for each high-risk function. Finish by checking test coverage and grouping findings by risk level with a merge recommendation.

Query the knowledge graph with the tests_for pattern for each high-risk changed function. For any untested changes, the review suggests specific test cases to add before merging.

It is a review approach that queries a graph of the codebase to trace execution flows and blast radius instead of reading files manually. The graph narrows scope but does not replace reading the actual implementation and its tests.

Start with a minimal-context query for your task and use minimal detail level on all graph calls, escalating to standard only when needed. The target is completing a review in five or fewer tool calls and under 800 output tokens.

Graph analysis narrows scope but cannot replace reading source code. Always read the implementation and its tests before changing code, since the graph shows relationships, not full logic.