The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to use the --dangerously-skip-permissions flag when invoking the agy CLI. This bypasses the application's internal authorization prompts, allowing the delegated agent to use all available tools without explicit user approval for each action.
[COMMAND_EXECUTION]: The workflow defaults to using --mode accept-edits, which grants the child agent the ability to modify the local filesystem and apply changes automatically without a human-in-the-loop review process.
[EXTERNAL_DOWNLOADS]: The skill recommends the installation of third-party system dependencies, including tmux via Homebrew on macOS and pywinpty via pip on Windows, to support interactive PTY sessions.
[COMMAND_EXECUTION]: The skill executes a Python script (orchestrator_supervisor.py) from a relative path (../orchestrator-cli/), creating a dependency on the presence and integrity of a separate skill's directory structure.
[PROMPT_INJECTION]: The skill processes arbitrary task descriptions and injects them into a high-autonomy delegation chain. This presents an indirect prompt injection surface where malicious data within a processed workspace could influence the delegated agent's behavior while safety constraints are disabled.

antigravity-cli