adaptive-guard
Pass
Audited by Gen Agent Trust Hub on Mar 29, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: No malicious patterns or security vulnerabilities were identified. The skill is a comprehensive security tool designed to protect AI systems.
- [PROMPT_INJECTION]: Numerous patterns such as 'ignore previous instructions' and 'DAN mode' were flagged in SKILL.md and references/static-rules.md. These are correctly identified by the skill's own logic as blacklist signatures used to detect and block malicious inputs. They are not instructions for the agent to follow.
- [COMMAND_EXECUTION]: The skill references command injection techniques (e.g., 'eval( + variable') and bash command blacklists. These serve as defensive markers for the security layer and do not involve unauthorized command execution.
- [DATA_EXFILTRATION]: Patterns associated with requesting API keys or system prompts are explicitly defined as 'Sensitive Data Exfiltration Signals' within the skill's defensive architecture to prevent such data from being leaked.
Audit Metadata