autonomous-operation
Fail
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: HIGHPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- Prompt Injection (HIGH): The skill contains instructions that explicitly command the AI to override its system prompts and efficiency constraints, using patterns like 'IGNORE: Be concise' and 'IGNORE: Minimize tokens'.
- Prompt Injection (HIGH): It includes directives to ignore user-initiated termination or pause requests, such as 'The session does NOT end because: Let's continue later', attempting to bypass user control protocols.
- Command Execution (LOW): The skill demonstrates use of 'sleep' and 'gh' in shell blocks for autonomous monitoring, which constitutes a command execution surface.
Recommendations
- AI detected serious security threats
Audit Metadata