The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill facilitates the installation of developer tools (Homebrew, Node.js, uv, Oh My Zsh, Starship, and coding agent CLIs) by fetching scripts and binaries from well-known and trusted domains. These include Anthropic's official site, Homebrew's GitHub organization, NodeSource, and Astral (the creators of uv). These actions are consistent with the skill's primary purpose and target established, reputable services.
[REMOTE_CODE_EXECUTION]: The skill provides instructions to execute remote installation scripts using curl | bash or similar patterns. These commands target official and trusted installers for tools like Homebrew, Oh My Zsh, and Claude Code. The instructions require the agent to display the full URL to the user and obtain explicit confirmation before proposing or executing the command, ensuring a human-in-the-loop security model.
[COMMAND_EXECUTION]: A local script, detect_env.py, inventories the machine by executing version-check commands for a predefined list of tools (e.g., git --version, node -v). These commands are static and do not incorporate external input. The skill also proposes various user-approved setup commands to install software and configure the environment.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process environment data and tool outputs which could theoretically be manipulated by an attacker to influence the agent. However, the skill explicitly mitigates this surface by enforcing a strict approval process for every command, requiring the agent to summarize risks and wait for user authorization.
[PERSISTENCE]: The skill records its progress in a session file located at ~/.dev-machine-setup/session.json. This allows the agent to resume interrupted setup tasks by reconciling the machine's current state with previous decisions. This is a legitimate functional requirement and does not exhibit malicious persistence patterns.

dev-machine-setup