multi-agent-code-review

Parallel Security, Performance, and Style Code Review

Code is parsed into an AST (functions: ["handleRequest", "validateInput", "processData", "sendResponse"], complexity score). Three reviewers run in parallel: security finds SQL_INJECTION (HIGH), performance finds N_PLUS_1_QUERY (HIGH), style finds INCONSISTENT_NAMING (LOW). A compiler merges all findings.

Workflow

code, language -> cr_parse_code
  -> FORK_JOIN(cr_security_review | cr_performance_review | cr_style_review)
  -> cr_compile_review

Workers

ParseCodeWorker (cr_parse_code) -- Returns AST with functions, imports, line count, complexity.

SecurityReviewWorker -- Finds {severity: "HIGH", type: "SQL_INJECTION"}.

PerformanceReviewWorker -- Finds {severity: "HIGH", type: "N_PLUS_1_QUERY"}.

StyleReviewWorker -- Finds {severity: "LOW", type: "INCONSISTENT_NAMING"}.

CompileReviewWorker -- Merges findings (null-safe via List.of() defaults).

Tests

38 tests cover parsing, all three review types, and compilation.

Further Reading

• RUNNING.md -- how to build and run this example