ai-output-humanizer
Pass
Audited by Gen Agent Trust Hub on Jul 7, 2026
Risk Level: SAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process user-supplied text and local files, which represents a potential surface for indirect prompt injection if the content contains malicious instructions. However, the risk is mitigated by explicit instructions to the agent to distinguish between commands and content. Additionally, the skill's restricted toolset (no network access) prevents data exfiltration or remote code execution via this vector.
- Ingestion points: User-provided text strings and local files accessed via the Read and Edit tools.
- Boundary markers: Explicit instructions to ignore instruction-like text within the content to be humanized.
- Capability inventory: Read, Write, Edit, Grep, Glob, and AskUserQuestion tools.
- Sanitization: The skill does not perform automated sanitization but relies on the LLM's adherence to the provided guardrails.
Audit Metadata