The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies extensively on executing shell commands and system utilities to perform its functions. It invokes du, lsof, ps, git, and osascript (AppleScript) through the disk_tidy.py script and via direct instructions to the agent in SKILL.md. While these are standard tools for its intended purpose, the wide range of executed commands increases the risk if the agent is manipulated into performing unintended operations.
[DATA_EXFILTRATION]: The disk_tidy.py script accesses and reports the command-line arguments of running processes by reading /proc/{pid}/cmdline (on Linux) or executing ps (on macOS). This functionality is designed to prevent the deletion of in-use folders, but it exposes full command strings to the agent. This may inadvertently reveal sensitive information, such as API tokens, session keys, or private configuration values that have been passed as CLI arguments to other running applications.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the local file system and process metadata, creating a surface for indirect injection attacks.
Ingestion points: Directory names, file paths, and process command-line arguments are read into the agent's context during the survey and merge phases (Workflow steps 2 and 4).
Boundary markers: The skill includes specific rules to mitigate accidental execution, such as conducting a read-only survey first and requiring explicit user confirmation for each item in the cleanup list.
Capability inventory: The skill possesses capabilities for directory deletion (shutil.rmtree), file system modification, and arbitrary command execution through multiple system binaries.
Sanitization: The disk_tidy.py script implements validation checks, ensuring that only specifically named folders (e.g., .next, .turbo, target) are eligible for deletion and verifying their git-ignore status before removal.

disk-tidy