The Agent Skills Directory

[EXTERNAL_DOWNLOADS] (SAFE): Documentation recommends installing standard, well-known libraries (browser-use, langchain-anthropic, playwright) from the official PyPI registry.
[COMMAND_EXECUTION] (SAFE): Example shell scripts perform routine tasks such as setting file permissions (chmod +x), installing packages, and managing local browser processes (taskkill.exe) for debugging purposes.
[DATA_EXFILTRATION] (SAFE): The documentation uses standard placeholders for API keys (e.g., 'sk-ant-...') and correctly recommends the use of environment variables for secret management.
[Indirect Prompt Injection] (SAFE): The skill describes an architecture for browser agents which, by design, ingests untrusted web data. While this creates a vulnerability surface for indirect prompt injection, the documentation is for the library itself and not a specific malicious implementation.
Ingestion points: The Agent class in api-reference.md is designed to navigate and extract data from external websites.
Boundary markers: Examples do not explicitly demonstrate prompt delimitation for ingested content.
Capability inventory: The agent has capabilities to click, type, and navigate as defined in the Controller actions.
Sanitization: The reference docs do not provide specific sanitization logic, as that is typically implementation-dependent.

browser-use-integration