code-review-graph

Reduce AI code review token usage by computing blast radii of changes in an AST graph instead of reading entire codebases, cutting token consumption by 6.8x to 49x on average. This approach provides only impacted files to the AI assistant.

Tree-sitter AST graph analysis parses your codebase structure into a SQLite database, enabling efficient structural queries. When you modify a function, it maps the blast radius to identify exactly which files are impacted by that specific change.

Code-review-graph is optimized for codebases over 500 files, especially for multi-file changes in monorepos or complex modules. Smaller projects may not see significant token reduction benefits from blast radius analysis.

You need Tree-sitter for parsing, SQLite for the graph database, and pymysql for database connectivity. The Skill uses minimal Python and MCP dependencies to maintain efficiency during structural analysis.

Use blast radius analysis when reviewing multi-file changes in monorepos or complex modules where providing full codebase context would exceed token limits. It identifies only impacted files, making it ideal for large-scale codebases over 500 files.