humanizer-zh

Pass

Audited by Gen Agent Trust Hub on Sep 5, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input for text editing.\n
  • Ingestion points: User-supplied text provided via prompts triggered by '中文润色' (Chinese polishing) or '审阅' (Review) keywords as defined in SKILL.md.\n
  • Boundary markers: SKILL.md contains explicit instructions to 'Keep code, identifiers, commands, formulas, URLs, link targets and metadata unchanged' and to 'preserve facts, numbers, and stances'.\n
  • Capability inventory: The skill only provides text transformation logic. It does not contain any code or shell commands capable of executing subprocesses, file operations, or network requests.\n
  • Sanitization: Instructions demand the preservation of technical symbols and strict adherence to provided facts, which mitigates the risk of the agent following instructions embedded in the input text.\n- [EXTERNAL_DOWNLOADS]: The documentation provides manual installation instructions via npx and git clone from external GitHub repositories.\n
  • Evidence: README.md contains npx skills add https://github.com/op7418/Humanizer-zh.git and git clone https://github.com/op7418/Humanizer-zh.git. These are intended for user setup and are not part of the skill's runtime execution logic.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 5, 2026, 08:31 PM
Security Audit — agent-trust-hub — humanizer-zh