Code reviews that come back with findings, not opinions.
Paste your goal and get a review prompt with the dimensions fixed, severity levels defined, and an output shape you can act on.
What the prompt you get back gets right
Review dimensions fixed
Correctness, security, performance and readability are named in the prompt, so the model can't quietly skip the one you care about.
Severity you can triage
Every finding is labelled — critical, major or minor — so you know what blocks the merge and what can wait.
Tied to the code it saw
Findings reference the exact function or lines they came from, instead of advice that could apply to any file.
No style nitpicks unless you ask
Formatting and naming preferences stay out of scope by default, so the review stays about behaviour.
From a rough note to a working prompt
Your note
“review this function before I merge it”
The prompt you get back
- Role
- a senior reviewer with the codebase context you paste in
- Scope
- what to check — correctness, security, performance, readability — and what to leave alone
- Severity
- critical / major / minor, each finding labelled with the reason it sits at that level
- Output
- location, the issue, why it matters, and a suggested fix you can apply
Every prompt is written for your exact request — this is the structure it comes back in, not a fixed template.
Common questions
Will it catch real security issues?
The prompt directs the model through the common classes — injection, unvalidated input, secrets in code, unsafe defaults — and asks it to mark anything it is not certain about as needing verification rather than asserting it.
Can I point it at a whole file or just a diff?
Either. The prompt states where the code goes and asks for findings tied to the lines it was given.
Try it with your own notes
One pass in, a structured prompt out — saved to your library so you can reuse it.
Engineer my promptOther tasks people engineer prompts for: