humanizer-zh
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSNO_CODEPROMPT_INJECTION
Full Analysis
- [External Downloads] (LOW): The skill is distributed via an untrusted GitHub repository (op7418/Humanizer-zh). Installation via 'npx skills add' or 'git clone' from unverified sources is a standard but noteworthy risk factor.
- [Indirect Prompt Injection] (LOW): The skill's primary function is to process and transform untrusted text (AI-generated content) provided by the user. 1. Ingestion points: The skill accepts user text via direct prompts and reads from files like 'article.md' mentioned in README.md. 2. Boundary markers: No explicit delimiters or warnings to ignore embedded instructions are documented in the README. 3. Capability inventory: The skill utilizes LLM capabilities for natural language rewriting and pattern identification. 4. Sanitization: There is no evidence of input sanitization or filtering to prevent malicious instructions embedded in the source text from influencing the agent's behavior.
- [No Code] (SAFE): The project structure suggests it is a prompt-only skill composed of markdown files (SKILL.md, README.md) with no executable scripts or binaries included.
Audit Metadata