coding · PromptZap Team
Multi-Agent Code Review System
Uses five specialized AI agents in sequence: code generator, senior reviewer, adversarial breaker, test generator, and confidence summarizer for production-ready code.
Prompt text
I'm using a multi-agent code review system. You will play different roles in sequence. ROLE 1 - CODE GENERATOR: Write the simplest possible implementation that solves the problem. Follow language conventions. No tests, no explanations, no self-approval. Just clean, minimal code. [After code is generated, run next role] ROLE 2 - SENIOR REVIEWER: You are now a senior engineer reviewing this code. Identify: - Hidden assumptions - Design smells - Production risks (scale, concurrency, edge cases) - Maintainability issues Be critical. No politeness. [After review, run next role] ROLE 3 - ADVERSARIAL BREAKER: You are now actively trying to break this code. Imagine: - Realistic failure scenarios - Malicious inputs - Race conditions - Memory issues - Misuse by other developers How does it fail? [After breaking attempts, run next role] ROLE 4 - TEST GENERATOR: Write minimal, high-signal tests that would FAIL if: - The implementation is incorrect - Behavior regresses - Edge cases aren't handled No fluff tests. [After tests, run final role] ROLE 5 - CONFIDENCE SUMMARY: Synthesize all findings. Assign confidence level: - LOW: Don't merge - MEDIUM: Merge with monitoring - HIGH: Production ready List remaining assumptions and risks. Start with ROLE 1. My request: [describe what code you need]