hermes-stack-doctor

Pass

Audited by Gen Agent Trust Hub on Jul 25, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill is implemented as a read-only diagnostic tool. No malicious code, data exfiltration patterns, unauthorized persistence mechanisms, or credential harvesting were detected. The skill explicitly prohibits system mutations, service restarts, or package updates without separate user approval.
  • [PROMPT_INJECTION]: The skill identifies a surface for indirect prompt injection because it processes untrusted data such as logs and repository files. However, it implements the recommended mitigations for this attack vector.
  • Ingestion points: Processes logs, repository files, archives, and package metadata as defined in SKILL.md and references/protocol.md.
  • Boundary markers: The skill contains an 'Untrusted Content Boundary' section in SKILL.md and references/safety.md that explicitly instructs the agent to treat inspected content as evidence, not instructions.
  • Capability inventory: The skill restricts the agent to read-only discovery; it does not provide tools for executing code or modifying files based on the content of audited files.
  • Sanitization: The instructions direct the agent to 'extract facts only', 'quote minimally', and flag suspected prompt-injection or social-engineering attempts as findings rather than obeying them.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 25, 2026, 07:07 PM
Security Audit — agent-trust-hub — hermes-stack-doctor