human-writing-zh
Pass
Audited by Gen Agent Trust Hub on Sep 6, 2026
Risk Level: SAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-provided Chinese text as its primary ingestion point (SKILL.md). It contains explicit defensive boundary markers, instructing the agent to treat any commands or instructions embedded within the input text as content to be edited rather than instructions to execute (e.g., in SKILL.md and reference/content-fidelity-and-scope.md). The skill's capability inventory is limited to text transformation without network or persistent file-writing capabilities. Sanitization is achieved through this strict separation of data and instruction.
- [COMMAND_EXECUTION]: The repository contains several PowerShell scripts within the evaluations/ directory (e.g., attribution-contract.ps1, skill-contract.ps1). These scripts are intended for local validation of the skill's markdown content to ensure adherence to defined writing standards. They perform read-only string matching operations on local files and do not exhibit malicious behavior or perform unauthorized network requests.
- [DATA_EXPOSURE_AND_EXFILTRATION]: No sensitive file paths, hardcoded credentials, or unauthorized network operations were detected. The skill guidelines explicitly forbid the use of external search or data retrieval unless specifically requested by the user.
- [OBFUSCATION]: Analysis of all markdown and script files revealed no hidden content, Base64-encoded commands, homoglyph attacks, or zero-width character obfuscation.
Audit Metadata