attacks

Prompt Injection Attack Library

40 categorized prompt-injection cases for testing LLM system-prompt security.

Category	Cases	Purpose
Direct Override	8	Explicitly replace/ignore instructions
Role Hijacking	6	Force a new persona or mode
Context Smuggling	6	Hide malicious instructions in documents/data
Encoding Tricks	5	Obfuscate attack payloads
Payload Splitting	4	Extract secrets piece by piece
System Extraction	6	Pull system prompts or internals
Multi-Turn	5	Simulate trust-building/escalation attacks

Run one suite without model calls:

proofloop run attacks/direct-override.yaml --dry-run

Run live against configured provider:

proofloop run attacks/direct-override.yaml