humanizer-zh-pro
Pass
Audited by Gen Agent Trust Hub on Nov 12, 2026
Risk Level: SAFEOBFUSCATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [OBFUSCATION]: Several files (README.md, SKILL.md, and detector-calibration.md) contain a sequence of zero-width characters and invisible separators at the start of the body text. The author's copyright notice in SKILL.md identifies these as embedded watermarks for plagiarism tracing. While declared as a protective measure, they constitute hidden data not visible in conventional text editing.
- Evidence: The hidden sequence
is embedded at the top of the content in multiple instruction files. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to accept and process arbitrary Chinese text from users to remove AI generation patterns. Because the agent is granted 'Write' and 'Edit' permissions on the file system, malicious instructions hidden within the user-provided 'AI-generated' text could potentially manipulate the agent into performing unauthorized file modifications.
- Ingestion points: Any user-supplied text provided to the agent for 'humanizing' or 'de-AI' processing.
- Boundary markers: The skill lacks explicit boundary markers or instructions for the agent to treat input text strictly as data rather than potential commands.
- Capability inventory: The skill is granted
Read,Write,Edit,Grep, andGlobtools, which provide significant control over the local project environment. - Sanitization: There are no instructions to sanitize or validate the content of the text provided for processing before the agent acts upon it.
Audit Metadata