Prompt Injection Test Kit
Shipping an LLM feature — a chatbot, RAG search, or a tool-using agent? Attackers will try to hijack it within minutes of launch. Test it first with real red-team prompts written by a security researcher.
Free: interactive attack library
12 sample prompt-injection attacks below. Pick your app type to see the ones that hit it hardest, copy a prompt, and paste it into your model or app to see how it responds. A hardened system should refuse or ignore every one.
Full Kit — 62 attacks across 9 categories
Everything in the sample, expanded and ready to run in CI:
- 62 hand-written attack prompts (JSON + CSV): direct override, prompt-leak, indirect/RAG injection, tool abuse, obfuscation, social engineering, data exfiltration, policy probes, format break-out. Each with severity + expected safe behavior.
- Python test runner — feed every attack to your model or your deployed app endpoint; auto-triages PASS / FAIL / REVIEW. Bring your own API key; nothing is sent to us.
- 20-point defense checklist — the concrete controls that actually fix what you find.
After payment, your download link is shown on the Stripe confirmation and emailed on your receipt. Bought it already? The link is in that receipt email.
Why this exists
Prompt injection is the top security risk for LLM applications: untrusted text — a user message, a retrieved document, a web page your agent reads — smuggles in instructions that override your system prompt, leak secrets, or trigger tool actions. You can't fix what you don't test. This kit gives you a concrete, repeatable test suite and the checklist to close the gaps. Test only systems you own or are authorized to test.