The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes various local development tools (git, npm, brew, docker, xcrun, etc.) to scan for and clean up disk space. While these are powerful commands, their use is restricted to a predefined whitelist in scripts/clean.sh and scripts/scan.sh.
[DYNAMIC_EXECUTION]: The clean.sh script dynamically constructs commands based on the cleanup plan. It mitigates risk by using a strict case statement for allowed command keys and applying regular expression validation to arguments such as UUIDs for simulators and package names for Android SDK components.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it ingests untrusted data from the local file system, including git logs, directory structures, and project metadata (e.g., CFBundleIdentifier from Plist files). This risk is addressed through the following measures:
Ingestion points: Git status/logs, directory listings via find and du, and Info.plist files via PlistBuddy.
Boundary markers: The skill requires an explicit human confirmation step for every cleanup tier and project, preventing autonomous execution on malicious data.
Capability inventory: The skill can move files to the Trash and execute whitelisted CLI tools, but it is explicitly forbidden from using sudo or rm -rf.
Sanitization: Arguments for dynamic commands are validated against strict character allowlists (regex).

mac-dev-storage-cleanup