The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the agent-browser CLI tool for navigation, interaction, and state management. This is the primary method of operation for the skill.
[EXTERNAL_DOWNLOADS]: The skill references and directs users to install the agent-browser utility from Vercel Labs' official GitHub repository (vercel-labs/agent-browser).
[PROMPT_INJECTION]: The skill manages an indirect prompt injection surface inherent to browser automation, where malicious instructions could be embedded in external web pages.
Ingestion points: Untrusted data enters the agent context via snapshot, screenshot, and get commands that read content from live websites.
Boundary markers: The documentation explicitly encourages the use of AGENT_BROWSER_CONTENT_BOUNDARIES to wrap external content with markers, helping the agent distinguish between data and instructions.
Capability inventory: The skill possesses capabilities to execute bash commands, write files, and run arbitrary JavaScript within the browser context using eval.
Sanitization: The skill provides hardening options, including domain allowlisting and an AGENT_BROWSER_ACTION_POLICY to restrict high-risk operations like eval, download, and network access.

agent-browser