The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill supports dynamic code execution through multiple commands.
The browser-use eval command executes arbitrary JavaScript within the context of the active web page.
The browser-use python command allows execution of arbitrary Python code with access to a persistent browser object.
The skill supports executing local Python files via browser-use python --file script.py.
[DATA_EXFILTRATION]: The skill facilitates access to highly sensitive user data through its 'real' browser mode.
Using browser-use --browser real allows the automation to use the user's actual Chrome profile, providing access to saved passwords, active login sessions, and browser cookies.
[COMMAND_EXECUTION]: The skill relies on a CLI tool (browser-use) invoked via Bash to perform all browser operations, which includes network navigation and file system writes for screenshots.
[PROMPT_INJECTION]: The skill possesses a significant attack surface for Indirect Prompt Injection (Category 8).
Ingestion points: The skill retrieves and processes untrusted content from external URLs via browser-use open, browser-use state, and browser-use extract in SKILL.md.
Boundary markers: The instructions lack delimiters or warnings to ignore instructions found within the processed web data.
Capability inventory: The agent has extensive capabilities to act on findings, including clicking elements, typing into fields, and executing code based on indices derived from untrusted page state.
Sanitization: There is no evidence of sanitization or validation of the data retrieved from web pages before it influences agent behavior.

Browser Use