The Agent Skills Directory

[COMMAND_EXECUTION]: The documentation references the use of CLI tools such as hermes and cua-driver for environment setup, system diagnostics, and skill pack management (e.g., hermes computer-use doctor, cua-driver skills install). These are presented as user-mediated configuration steps.
[EXTERNAL_DOWNLOADS]: The skill references the trycua/cua repository on GitHub for its underlying driver and provides a command to install additional platform-specific documentation and configuration files.
[PROMPT_INJECTION]: The skill explicitly identifies the risk of indirect prompt injection from on-screen content. It provides a hard rule for the agent to treat the user's original prompt as the only source of truth and to ignore any instructions found within screenshots or application windows.
[DATA_EXFILTRATION]: While the skill enables reading screen content and typing text, it contains strict prohibitions against the agent interacting with or typing sensitive data, including passwords, API keys, credit card numbers, or 2FA challenges.
[COMMAND_EXECUTION]: The type action implementation includes a security blocklist that prevents the agent from typing dangerous shell command patterns such as piped remote script execution or destructive system commands.

computer-use