The Agent Skills Directory

Sensitive Data Protection: The instructions include a specific prohibition against reading or displaying the .env file, using silent grep commands to verify the presence of an API key without exposing its content. This is a recommended security practice for managing credentials.
Environment Verification: The skill performs checks for required dependencies like playwright and computer_use using inline Python commands. These checks are limited to validating the installation and do not perform unexpected operations.
Controlled Tool Execution: The initial demonstration task is restricted to a headless browser environment by setting CU_ENABLE_COMPUTER_USE_TOOLS=false. This design ensures the agent cannot interact with the host system's physical peripherals until the user explicitly grants permission and changes the execution mode.
Local Development Interfaces: The skill utilizes standard frameworks such as Streamlit and Uvicorn to run a trajectory viewer and a tool panel locally. These components are intended for developer debugging and do not initiate external network connections to untrusted third-party services.

first-run