The Agent Skills Directory

[COMMAND_EXECUTION]: The browser-use python command allows for the execution of arbitrary Python statements to interact with the browser's raw Chrome DevTools Protocol (CDP) sessions, providing low-level access to browser internals.
[COMMAND_EXECUTION]: The browser-use eval command executes arbitrary JavaScript code within the context of the active browser tab, allowing for manipulation of web page content and behavior.
[DATA_EXFILTRATION]: The skill includes functionality to export browser cookies to a local JSON file via browser-use cookies export, which can be used to harvest active session tokens.
[DATA_EXFILTRATION]: The browser-use tunnel command utilizes Cloudflare tunnels to make local network ports accessible via a public URL, creating a potential vector for exfiltrating local data or exposing internal services.
[CREDENTIALS_UNSAFE]: The skill can connect to a user's local Chrome instance or specific profiles using browser-use connect or the --profile flag, granting the agent access to the user's stored passwords, browsing history, and active login sessions.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection. It ingests untrusted data from web pages via commands like browser-use state and get text (SKILL.md) without defined boundary markers or sanitization. This is particularly risky given the skill's capabilities, such as arbitrary code execution and the ability to export sensitive data like cookies.
[EXTERNAL_DOWNLOADS]: The browser-use profile update command downloads and updates the profile-use binary from the vendor's infrastructure to manage browser profiles.

browser-use