The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Downloads browser binaries from Chrome for Testing (Google) and fetches updates for the tool from its official Vercel-associated repository.
[COMMAND_EXECUTION]: Provides an eval command to execute arbitrary JavaScript code within the browser context. A specific flag -b is included to support the execution of Base64-encoded scripts, which is a common technique for obfuscating code intent.
[CREDENTIALS_UNSAFE]: Includes functionality to save, store, and manage encrypted credentials (usernames and passwords) within a local 'Auth Vault'.
[DATA_EXFILTRATION]: Features stream and dashboard commands that open local network ports for WebSocket communication and browser monitoring, which could be leveraged for unauthorized data access or session hijacking if not properly secured.
[PROMPT_INJECTION]: The agent-browser read and snapshot commands ingest and process external web content. This functionality serves as a potential vector for indirect prompt injection, as malicious instructions embedded in a web page could attempt to influence the agent's behavior.
[REMOTE_CODE_EXECUTION]: The upgrade command facilitates the download and execution of code from a remote source to update the tool's functionality.

agent-browser