agent-security
Pass
Audited by Gen Agent Trust Hub on Nov 2, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The file references/examples.md contains injection-like strings such as 'ignore prior instructions', but these are explicitly presented as examples of malicious behavior that the agent must identify and reject. The skill's purpose is to prevent prompt injection, not facilitate it.
- [DATA_EXFILTRATION]: The instructions mandate that the agent must not read or disclose sensitive data such as SSH keys, .env secrets, or environment variables to satisfy an external source's instruction. This proactively protects against credential harvesting.
- [REMOTE_CODE_EXECUTION]: The skill explicitly forbids piping remote content directly into an interpreter and requires that any external scripts be saved and inspected before use, prioritizing safety over direct execution.
- [SAFE]: No obfuscation, persistence, or malicious command execution patterns were identified. The skill acts as a set of security constraints for the agent.
Audit Metadata