prompt-injection-defense
Pass
Audited by Gen Agent Trust Hub on Mar 31, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill documentation includes strings commonly associated with injection attacks, such as "Ignore all previous instructions," "You are now in developer mode," and "reveal your system prompt." These are provided exclusively as data for testing defensive filters and as examples of patterns that the provided Python scripts are designed to detect.
- [EXTERNAL_DOWNLOADS]: References and demonstrates the integration of well-known, legitimate security tools and libraries, including NVIDIA's Garak (LLM vulnerability scanner) and Microsoft's PyRIT (Red Teaming Orchestrator).
- [DATA_EXFILTRATION]: Provides defensive code implementation for output filtering using regular expressions. These patterns are used to identify and mask sensitive data like API keys and passwords from model outputs, serving as a security hardening feature.
- [SAFE]: The skill uses Base64 and URL encoding in its examples to demonstrate how attackers might attempt to bypass filters. These encoded strings are part of a defined test suite for security validation and do not represent an attempt to obfuscate malicious intent within the skill itself.
Audit Metadata