Uses linters, tests, and repo queries via function calling / MCP. The LLM mainly interprets evidence and suggests minimal diffs. Quality-first approach grounded in real data.
Execute ESLint, Pylint, or language-specific linters
Execute test suite for affected files
Semantic search across repository
Check dependency graph and versions
Deep diff parsing and AST comparison
Tools are exposed via MCP (Model Context Protocol) for standardized integration. This allows the reviewer agent to dynamically discover and invoke tools based on the PR context, without hardcoding tool definitions.
Quality-critical codebases where accuracy matters more than speed