humanizer
Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-supplied text and file content to identify and rewrite AI writing patterns, creating a vulnerability surface for indirect prompt injection.\n
- Ingestion points: Content provided directly in chat or read from local files during 'File mode' (as specified in SKILL.md).\n
- Boundary markers: The skill does not define or use explicit delimiters (such as unique tokens or XML-style tags) to isolate the instructions from the untrusted data being processed.\n
- Capability inventory: In 'File mode', the agent is instructed to 'rewrite the file in place', which involves both file-read and file-write operations (SKILL.md).\n
- Sanitization: No explicit sanitization or filtering of external content is implemented before it is processed by the agent.\n- [COMMAND_EXECUTION]: The documentation suggests the execution of system commands for setup and maintenance.\n
- Evidence: README.md and AGENTS.md provide instructions to run 'npx skills add' and a local validation script 'python3 scripts/validate-package.py'.\n- [EXTERNAL_DOWNLOADS]: The skill references external sources for installation and documentation.\n
- Evidence: README.md and UPSTREAM.md link to repositories on GitHub (blader/humanizer) and vendor-managed resources (light-cheng/skills-3rdParty).
Audit Metadata