human-ai
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-provided text through a multi-step editorial pipeline. While it establishes factual guardrails, it lacks explicit instructions for the agent to use robust boundary markers (such as XML tags or unique delimiters) when handling user input. This creates a surface where malicious instructions embedded in the input text could attempt to override the agent's behavior. Evidence found in the humanization process (Steps 0-6) in SKILL.md. Capability inventory includes text rewriting and metric calculation via the local scripts/measure.py script, with no network or file-write permissions requested. No specific sanitization of input text is provided.
- [SAFE]: All external links and installation references, such as to the author's GitHub repository or well-known documentation sites like Wikipedia and TypeSafe AI, are to trusted or established services. These references are used for research and model selection and do not introduce security risks.
Audit Metadata