The original title is: "Prompt Injection Tester: 5 attack classes for auditing system prompt resilience"

Short summary

The author built a free Prompt Injection Tester that runs five attack classes — instruction override, prompt extraction, delimiter escape, role-play, and indirect injection — against a pasted system prompt and returns a resilience score with model responses as evidence. The tool runs locally so the system prompt never leaves your hardware. The key insight is that indirect injection via RAG or browsing tools is the most dangerous and often overlooked vector.

• •Free tool runs 5 attack classes against your system prompt: instruction override, prompt extraction, delimiter escape, role-play, indirect injection
• •Runs on your own hardware — system prompt is never sent to a third party
• •Indirect injection via RAG or browsing tools is the biggest blind spot for most teams

Generated with AI, which can make mistakes.