The Agent Skills Directory

[COMMAND_EXECUTION]: The browser-use python command (documented in references/cdp-python.md) enables the execution of arbitrary Python statements, providing a direct interface for system-level code execution in the agent's environment.
[REMOTE_CODE_EXECUTION]: The browser-use eval command allows for the execution of arbitrary JavaScript within the browser context, which could be exploited if the browser interacts with malicious or compromised web pages.
[DATA_EXFILTRATION]: The skill facilitates the extraction of sensitive browser information, including cookies via browser-use cookies export and session data from user Chrome profiles using the --profile flag.
[CREDENTIALS_UNSAFE]: Connection to the user's primary browser through browser-use connect exposes all logged-in accounts and private browser data to the skill's automation logic.
[EXTERNAL_DOWNLOADS]: The browser-use profile update command initiates a download of the profile-use binary from a remote source, presenting a risk of unverifiable code execution.
[PROMPT_INJECTION]: The skill presents a large attack surface for indirect prompt injection. Untrusted data enters the context through commands like browser-use state and browser-use get text (Ingestion points). There are no visible boundary markers or instructions to ignore embedded commands (Boundary markers). The skill possesses high-privilege capabilities including code execution, cookie extraction, and file access (Capability inventory), and there is no specified sanitization of retrieved web content (Sanitization).

browser-use