Agentic Code Review Workflow

Code review is a perfect multi-agent task — different concerns are best handled by specialised models with focused context. This workflow implements a three-agent pipeline you can run locally with Claude Code or integrate into your CI. **Agent architecture:** **SecurityScannerAgent** — Analyses the diff for OWASP Top 10 patterns, hardcoded secrets, unsafe eval usage, unvalidated inputs, and insecure dependency references. Outputs structured JSON with severity, file, line, and remediation suggestion. **StyleCheckerAgent** — Reviews code complexity (cyclomatic complexity estimation), naming conventions, function length, duplicate logic, and comment quality. Flags code that will be hard to maintain in six months. **CoverageAnalyserAgent** — Examines the diff and existing test files to identify untested code paths, missing edge case tests, and integration gaps. Suggests specific test cases with pseudocode. **CoordinatorAgent** — Takes all three agent reports as input and produces a final review document in Markdown with an executive summary, prioritised action items, and a go/no-go recommendation. **Integration options:** GitHub Actions workflow file included for running on every PR. Also works as a standalone Claude Code slash command (`/review`) you can invoke during development. Requires Claude API access (or Claude Code) and Node.js 20+.