Adversarial code review.

A hostile pass over a PR, file, or project — one prompt per angle, aimed at the failures that slip past normal review. Paste each into your coding assistant with the code in front of it.

How to use — fill the fields below once and they drop into every prompt. Copy a prompt into your LLM or coding assistant, with the diff pasted or the repo open, then read the output critically: models invent plausible bugs, so reproduce before you believe. Tick the box once you've run a prompt, skip any category that can't apply, and export the report when you're done.