The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits a vulnerability to indirect prompt injection (Category 8). It ingests untrusted text for processing and relies on an external configuration file (STYLE.md) that it treats as "authoritative" and capable of overriding its default logic. This configuration can be influenced or created by user input via "Init Mode" or "Update Mode".\n- Ingestion points: Processes arbitrary user-supplied text for humanization and reads the STYLE.md file from several paths, including user-defined environment variables and configuration directories.\n- Boundary markers: None. The instructions do not specify delimiters or provide warnings for the agent to ignore embedded instructions in the text being humanized.\n- Capability inventory: The skill has access to Read, Write, Edit, Grep, and Glob tools, which allow for substantial file system interaction.\n- Sanitization: None. The skill does not implement validation or escaping for the input text or the style rules before processing.

humanizer