The Agent Skills Directory

[Indirect Prompt Injection] (HIGH): The skill exhibits a significant attack surface for indirect injection because it is designed to ingest and modify local files based on user-provided input.\n
Ingestion points: validate_python.py and validate_yaml.py accept arbitrary file paths via command-line arguments.\n
Boundary markers: No delimiters or 'ignore instructions' warnings are present; the skill treats all file content as valid data for analysis and transformation.\n
Capability inventory: The skill utilizes Path.write_text to modify and overwrite files on the filesystem, including injecting new code blocks.\n
Sanitization: The tool lacks path validation or sandboxing, meaning it does not verify if a path is within the intended skill workspace before performing read/write operations.\n- [Command Execution] (MEDIUM): The 'auto-fix' logic automatically performs writes to the host filesystem. If an agent is directed to 'fix' a sensitive system file (e.g., .bashrc, .ssh/config, or environment files), the skill will inject Python docstrings or boilerplate functions, potentially corrupting the file or disabling system services.

skill-validator