The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides native commands for arbitrary code execution. browser-use python "statement" enables the execution of arbitrary Python code with persistent state across calls (File: references/cdp-python.md). browser-use eval "js code" allows for the execution of arbitrary JavaScript within the active browser context (File: SKILL.md).
[DATA_EXFILTRATION]: Facilitates the extraction of sensitive session data and user information. browser-use cookies export and browser-use cookies get allow for the programmatic retrieval of session cookies. The Python CDP bridge provides access to the Network.getCookies method for detailed cookie extraction (File: references/cdp-python.md). browser-use connect enables the agent to interact with the user's main Chrome profile, providing access to saved logins and authenticated sessions (File: SKILL.md). browser-use screenshot can be used to capture sensitive information displayed on web pages.
[COMMAND_EXECUTION]: Provides a comprehensive CLI for system and browser management. It enables management of background daemons and browser sessions via the browser-use command. browser-use tunnel utilizes Cloudflare's service to establish network tunnels and expose local ports to the internet (File: SKILL.md).
[CREDENTIALS_UNSAFE]: Manages sensitive authentication data in local storage. browser-use cloud login stores API keys in a local configuration file located at ~/.browser-use/config.json (File: SKILL.md).
[EXTERNAL_DOWNLOADS]: Fetches external components for updates. browser-use profile update downloads and updates the profile-use binary from the vendor's official resources (File: SKILL.md).
[PROMPT_INJECTION]: The skill's architecture presents a surface for instructions embedded in external data (Indirect Prompt Injection). Ingestion points: Untrusted data enters the agent context via browser-use state, browser-use get html, and browser-use get text (File: SKILL.md). Boundary markers: No delimiters or safety instructions are present to distinguish between web content and agent commands. Capability inventory: The toolset includes arbitrary code execution (python, eval), cookie management, and network tunneling. Sanitization: There is no evidence of input validation or content filtering for data retrieved from the web.

browser-use