humanize
Pass
Audited by Gen Agent Trust Hub on Sep 6, 2026
Risk Level: SAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external prose (articles, READMEs, announcements) which constitutes untrusted input. However, it contains strong internal guardrails including explicit instructions to 'Never fabricate' facts or 'Never inject' voice/humor not present in the source. There are no associated dangerous capabilities like network operations or shell execution that could be triggered by data content.
- [EXTERNAL_DOWNLOADS]: The skill references multiple external research papers and datasets (from arxiv.org, Wikipedia, and GitHub). These are documented in the
.dev/reference-inventory.jsonandreferences/baselines.mdfiles as citations for its stylometric thresholds. None of these URLs are used for code execution or automated downloads. - [OBFUSCATION]: The documentation explicitly mentions 'Machine residue' such as zero-width spaces (U+200B, U+200D) and homoglyphs. These are treated as artifacts to be identified and removed from prose drafts rather than being used maliciously within the skill itself.
- [COMMAND_EXECUTION]: No shell commands or dynamic context injection patterns were found in the skill's instructions or metadata.
Audit Metadata