humanizer
Pass
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external text and files, representing an ingestion point for potentially untrusted data. In 'File mode', the agent reads from and writes to local files. While this creates an indirect prompt injection surface where malicious instructions in source text could attempt to influence the agent, the skill includes explicit boundary markers directing the agent to ignore code, data, and metadata, effectively mitigating the risk. Evidence: 1. Ingestion points: User-provided text and local file paths (SKILL.md). 2. Boundary markers: Explicit instructions to ignore non-prose elements (SKILL.md). 3. Capability inventory: File read and write operations for humanizing prose (SKILL.md). 4. Sanitization: None beyond stylistic transformation rules.
Audit Metadata