testing · PromptZap Team

Adversarial Input Fuzzer

Tries to break your form or API on purpose with the 15 nastiest realistic inputs, grouped into broken / unverified / test-manually.

Framework

standard

Model targets

claude, chatgpt, gemini

Prompt text

Pretend you're trying to break this form/API/feature on purpose: [describe it or paste code].

Generate the 15 nastiest inputs a real (or malicious) user might submit, and for each explain what you expect to happen vs. what you're worried might actually happen. Include things like: extremely long strings, script tags, SQL-looking strings, unicode/emoji, whitespace-only input, duplicate/rapid-fire submissions, wrong content-type, and expired/tampered tokens.

Then group your findings into: "definitely broken," "probably fine but unverified," and "need to test manually to be sure."