Prompt Injection Tester

Building an LLM app? Paste your system prompt below. This tool runs a defensive hardening checklist against it, scores it out of 100, and shows which classic injection categories your prompt is not yet defended against, so you can fix the gaps before attackers find them.

Results will appear here as soon as you paste a prompt.

Know the attack categories you are defending against

These are the ten classic prompt injection categories every LLM app builder should test against, described defensively: what the category is, and which checklist gap it exploits. After you paste a prompt above, each card shows whether your prompt currently has coverage for it. Use a proper red-team evaluation before shipping; this page is the study guide, not the exam.

Hardened prompt-sections template

A starting skeleton that covers every check on this page. Replace the placeholders with your app's specifics and keep the sections that apply.


        
      


      

        How the score works: this is a heuristic keyword-and-pattern scan of your prompt text, weighted by how much each defense matters in practice (resisting override attempts and delimiting untrusted input are weighted highest). A high score means your prompt states the right rules; it does not guarantee the model will always follow them. Treat this as a pre-flight checklist, then validate with real adversarial testing against your deployed app. Everything runs locally in your browser; your prompt is never sent anywhere.
      


      
Free tool from AI School - no signup, runs in your browser.