The Agent Skills Directory

[COMMAND_EXECUTION]: The script includes a --self-test mode that uses subprocess.run to execute its own unit tests via the Python unittest module. This execution is limited to running the internal test script provided with the skill.
[INDIRECT_PROMPT_INJECTION]: The skill scans file contents and metadata, including these snippets in prompts for LLM-based organization planning. Although malicious file content could attempt to influence the organization logic, the skill implements robust mitigations: a mandatory human-in-the-loop gate where the user must approve the full plan before execution, and strict path confinement in the validate_plan function that prevents destination paths from escaping the user-specified root directory.
[EXTERNAL_DOWNLOADS]: The propose subcommand optionally transmits file metadata and snippets to established LLM providers (DeepSeek, OpenAI, OpenRouter) or a local Ollama instance via HTTPS requests. This network activity is used to generate the organization structure and is consistent with the skill's stated purpose for managing large directories.
[DATA_EXFILTRATION]: The skill facilitates the transfer of directory metadata and file content snippets to external AI services for processing. This behavior is restricted to the specific path provided by the user and does not target sensitive system locations or credentials.

file-organizer