The Prompt

You are a staff engineer designing a system prompt for an AI coding assistant… [Full prompt as specified in frontmatter]

Why This Prompt Works

Severity taxonomy. The BLOCKER/WARNING/SUGGESTION/PRAISE categories prevent the AI from treating a missing semicolon the same as an SQL injection vulnerability. Baseline code review prompts produce flat lists where critical security issues are buried between style nits.

Scope guard. “Never suggest refactors that would expand the PR scope” is the most important constraint for real-world usage. Without it, AI reviewers suggest rewriting entire files when you changed two lines — this kills developer trust in the tool.

Cross-model portability. Requiring no model-specific syntax means the system prompt works if you switch from Claude to GPT-4 next quarter. This is practical engineering, not vendor lock-in.

Model Outputs

Claude 4 Sonnet

Gemini 2.5 Pro

DeepSeek V3

Grok 3

Without The Prompt (Baseline)

Task given: “Write a system prompt for an AI code review assistant”

Claude 4 Sonnet (Baseline)

Verdict

Try It Yourself

Copy the prompt above and try it in your preferred AI tool. For more system prompts, see our System Prompts shelf. Looking for role-specific prompts? Check out The AI for Work for professional workflows.

Code Review System Prompt

## what it does