solo-humanize
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from files or user input to perform text humanization. This creates an indirect prompt injection surface as the agent is instructed to process and rewrite external content while possessing file-writing capabilities.\n
- Ingestion points: File paths or pasted text provided via arguments (SKILL.md).\n
- Boundary markers: No explicit delimiters or safety instructions for input processing are defined.\n
- Capability inventory: The skill utilizes Read, Write, and Edit tools to modify local documents.\n
- Sanitization: Content is rephrased according to a pattern catalog, but no formal security sanitization is performed on the input data.\n- [SAFE]: No evidence of hardcoded credentials, remote code execution, or network communication was found. The skill's operations are confined to the local filesystem and text processing.
Audit Metadata