ja-humanizer
Pass
Audited by Gen Agent Trust Hub on Oct 28, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill is a legitimate tool for 'humanizing' AI-generated Japanese text. It works by applying linguistic patterns and user-provided voice samples to detect 'unnatural' AI rhythm or word choice.
- [SAFE]: The execution flow involves running a JavaScript-based checker (
ja-humanizer-check.mjs) locally to identify linguistic patterns. The logic is transparent and focuses on text analysis. - [SAFE]: All external URLs and references (e.g., GitHub, Qiita analysis by Sakasegawa) are to trusted or well-known technical resources used for the skill's legitimate purpose of linguistic benchmarking.
- [PROMPT_INJECTION]: The detector flagged a 'Policy Bypass' in
references/voice-template/README.md. Upon review, this is a FALSE POSITIVE. The text describes how voice samples are material, not instructions, and explains that sentences like 'please do this' inside a sample are data, not commands. This is a safety instruction to the agent, not a malicious bypass. - [DATA_EXPOSURE]: The skill accesses local files like
~/.agents/voice/to read the user's voice samples. This is a core part of its documented functionality (voice matching) and there is no evidence of exfiltration to external servers.
Audit Metadata