The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The browser-use python command allows for the execution of arbitrary Python code on the host system, enabling unauthorized host access and control.
[REMOTE_CODE_EXECUTION]: The browser-use eval command enables arbitrary JavaScript execution within the browser context, allowing for manipulation of the DOM and potential data theft.
[DATA_EXFILTRATION]: The browser-use profile sync command facilitates the extraction of sensitive session cookies from local Chrome profiles and uploads them to a cloud-based service.
[CREDENTIALS_UNSAFE]: Use of the --browser real flag grants the agent access to the user's primary browser profile, including all active login sessions and authentication tokens.
[PROMPT_INJECTION]: The skill is highly susceptible to indirect prompt injection. Ingestion points: browser-use open, state, and get html ingest untrusted content from arbitrary websites. Boundary markers: There are no mechanisms defined to prevent the agent from following instructions embedded in web content. Capability inventory: The agent possesses high-impact tools including host-level code execution (python), file exfiltration (profile sync), and authenticated browser access (--browser real). Sanitization: Web content is not sanitized before being processed by the agent.
[COMMAND_EXECUTION]: The skill relies on the Bash tool to execute CLI commands, providing a significant attack surface for command-line based exploits.
[EXTERNAL_DOWNLOADS]: The installation instructions involve downloading the browser-use package and its Chromium dependencies from external, third-party sources.

browser-use