The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands to modify system settings and search for file references. It executes defaults write com.apple.screencapture location and killall SystemUIServer on macOS to change the default screenshot directory to the skill's inbox. It utilizes rg (ripgrep) to search for absolute file paths within configuration and shell profile files, including ~/.zshrc, ~/.config, and .env files, to identify potential broken links caused by moving files. The test suite (scripts/test_file_tidy.py) employs subprocess.run to execute the main engine during validation.
[CREDENTIALS_UNSAFE]: The skill is designed to manage and organize sensitive files. It identifies and categorizes files containing identity documents, passwords, tokens, keys, and credentials into a designated 'Sensitive Zone' (e.g., ~/Files/99_sensitive). It scans highly sensitive paths such as ~/.zshrc, ~/.config, and various .env files to find references to moved files. The instructions explicitly forbid the agent from opening or printing the contents of files identified as sensitive to the interaction transcript.
[INDIRECT_PROMPT_INJECTION]: The skill processes arbitrary data from user-controlled directories, creating a potential attack surface. Ingestion points: The agent is instructed to scan the Downloads folder, the Desktop, and other user-specified directories to classify new files. Boundary markers: The skill provides a logical boundary by instructing the agent to never open or print files destined for the 'Sensitive Zone'. Capability inventory: The skill possesses extensive local file manipulation capabilities, including os.rename, os.link, shutil.copy2, and shutil.rmtree via the scripts/file_tidy.py engine. Sanitization: The file_tidy.py script implements rigorous validation, including a protected_reason function that refuses to modify system directories, home folders, or application data, and a snapshot function for content-hash verification of all moves.

file-tidy