ai-security
Pass
Audited by Gen Agent Trust Hub on Jun 16, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill contains numerous examples of prompt injection and jailbreak signatures (e.g., 'disregard prior context', 'DAN mode', 'repeat your system prompt'). However, these are documented explicitly as educational examples for the tool's signature matching engine and for security testing purposes. They are not intended to override the agent's own safety guidelines but to describe what the security professional should look for.
- [REMOTE_CODE_EXECUTION]: The skill mentions a Python script
scripts/ai_threat_scanner.py. While this script is not provided in the snippet, the documented usage shows it performing local static signature matching and JSON processing, which is standard behavior for a security scanner. - [DATA_EXPOSURE]: The skill discusses data exfiltration and model inversion as risks to be assessed. It provides guidance on how to monitor for these attacks, such as detecting systematic membership inference queries. No evidence of hardcoded credentials or unauthorized data access was found in the provided files.
- [COMMAND_EXECUTION]: The skill includes shell command examples for running the security scanner. These are standard CLI invocations for the provided tool and do not involve privilege escalation or unauthorized persistence mechanisms.
Audit Metadata