humanize-en
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external prose and configuration files, which are ingestion points for potentially untrusted data. \n
- Ingestion points: The skill accepts prose as inline text, file paths, and brand-voice specifications through command-line arguments. \n
- Boundary markers: The
mask_protected_regionsfunction inscripts/prescan.pyensures that non-prose elements like YAML frontmatter, URLs, and fenced code blocks are excluded from the scanning process. \n - Capability inventory: The skill is permitted to read and write files (using
ReadandEdit) and execute local Python scripts (Bash). \n - Sanitization: The transformation process is focused on stylistic rewriting, and the skill implements a final approval gate requiring the user to explicitly confirm changes before they are applied. \n- [COMMAND_EXECUTION]: The skill executes local Python scripts (
prescan.py,brand_prescan.py,validate.py) for deterministic pattern matching and regression testing. These scripts use standard Python libraries and are confined to the task of linguistic analysis.
Audit Metadata