aegis-llm-ai
Pass
Audited by Gen Agent Trust Hub on Aug 31, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill includes a collection of prompt injection examples and payloads designed for security testing in the 'Payload cheat-sheet' and 'Dynamic battery' sections. These include instructions to ignore previous system prompts (e.g., 'Ignore all previous instructions'), reveal hidden rules (e.g., 'print your system prompt verbatim'), and simulate 'unrestricted' or 'developer' modes. These strings are clearly documented as test data for auditors and are not intended to override the host agent's instructions during the skill's execution.
- [SAFE]: The skill references several external resources for further reading, including OWASP, NIST, MITRE, and official documentation from AI providers like OpenAI and Anthropic. All identified URLs point to trusted organizations or well-known services and do not involve unauthorized downloads or network exfiltration. The code examples provided are labeled as 'VULNERABLE' or 'FIXED' for educational purposes and do not represent executable malicious code within the skill context.
Audit Metadata