The Agent Skills Directory

Credential Protection: The skill contains explicit instructions to avoid reading or displaying sensitive files like .env. It uses silent checks (grep -q) to verify the presence of API keys without exposing them in the agent's context or conversation history.
Restricted Initial Capabilities: For the first run, the skill purposefully disables computer-interaction tools (mouse and keyboard) via the CU_ENABLE_COMPUTER_USE_TOOLS=false flag, limiting the agent to a browser-only environment to ensure a controlled and safe user experience.
Standard Environment Verification: The skill performs routine checks for required libraries and local configuration files using python -c, which is a standard pattern for ensuring a functional development environment.
Indirect Prompt Injection Surface: The skill involves the agent navigating to external websites (Wikipedia). While this introduces a potential surface for indirect prompt injection, the risk is managed by the restricted toolset and the specific, narrow scope of the initial task.
Ingestion points: The agent reads data from Wikipedia during the browser task in SKILL.md.
Boundary markers: The instructions designate the activity as a "safe browser-only task."
Capability inventory: The skill uses python -m computer_use, streamlit, and uvicorn for browser automation and UI display.
Sanitization: No specific sanitization of the browser content is described, which is common for general-purpose browsing tasks.

first-run