humanize
Pass
Audited by Gen Agent Trust Hub on Apr 7, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: No prompt injection or behavior override patterns detected. The instructions strictly define editing constraints (e.g., prohibition of em dashes and specific AI phrases) without attempting to bypass platform safety filters.
- [DATA_EXFILTRATION]: No network operations (curl, wget, etc.) or access to sensitive file paths (~/.ssh, ~/.aws, .env) were found. Data access is restricted to the local project's content and context files.
- [REMOTE_CODE_EXECUTION]: The skill does not install external dependencies or execute remote scripts. It operates entirely through prompt-based sub-agents within the platform environment.
- [PROMPT_INJECTION]: While the skill processes user-provided text (Indirect Prompt Injection surface), it lacks dangerous execution capabilities like 'eval', 'exec', or shell subprocesses. The presence of a 'Critic' agent that audits output against 36 patterns and absolute prohibitions acts as a significant mitigating factor for input poisoning.
Audit Metadata