The Agent Skills Directory

COMMAND_EXECUTION (LOW): The skill utilizes a local Python script scripts/format_fixer.py to process text formatting. The script is called via a shell command involving input redirection, which is a standard but sensitive operation requiring the agent to correctly handle file paths to avoid command injection.- PROMPT_INJECTION (LOW): The skill ingests untrusted user content (drafts) for processing by the AI. This creates a surface for Indirect Prompt Injection where malicious instructions in the draft could attempt to override the AI's conservative editing rules. The risk is mitigated by the diff-based approval workflow described in the workflow.- DATA_EXPOSURE (SAFE): The skill reads specific local files provided by the user and writes to a new filename with the polished- prefix. It does not appear to access sensitive system directories or environment variables.- EXTERNAL_DOWNLOADS (SAFE): No external dependencies or remote script downloads were detected. The formatting logic is entirely local.

polisher