The Agent Skills Directory

[COMMAND_EXECUTION] (LOW): The skill provides detailed patterns for executing powerful system commands including osascript (AppleScript), defaults (system preferences), and launchctl (service management). While these facilitate deep system modification, they are provided as instructional templates for the user.
[EXTERNAL_DOWNLOADS] (LOW): Reference materials include usage of curl and Homebrew (brew install) for fetching remote data and installing third-party packages. No instances of 'curl | bash' or other dangerous remote execution patterns were found.
[CREDENTIALS_UNSAFE] (LOW): The references/macos-commands.md file documents the security utility for interacting with the macOS Keychain. This includes commands to 'read password' and 'add to keychain'. While instructional, an agent acting on these commands could access sensitive credentials if directed by a user or an indirect injection.
[Indirect Prompt Injection] (LOW): The skill has a significant attack surface as it is designed to interact with system data that could be attacker-controlled.
Ingestion points: Reads data via pbpaste (clipboard), mdfind (Spotlight search/file contents), and cat (file reads).
Boundary markers: None present in the provided shell script templates.
Capability inventory: Includes sudo for privilege escalation, launchd for persistence, and osascript for arbitrary execution.
Sanitization: Shell templates follow defensive practices (e.g., set -u, variable quoting) but do not include specific sanitization for LLM prompt injection targets.

unix-macos-engineer