The Agent Skills Directory

[COMMAND_EXECUTION]: The skill operates by executing the agent-browser CLI to perform browser actions such as navigation, interaction, and session management.
[DYNAMIC_EXECUTION]: Includes an eval command that allows the execution of arbitrary JavaScript within the browser context. Documentation provides examples using base64 encoding to ensure reliable code execution and mentions standard automation use cases.
[EXTERNAL_DOWNLOADS]: Instructions guide the user to install the core CLI via official package managers (Homebrew, NPM) and support extending functionality via plugins from GitHub or NPM. The official source repository is hosted under Vercel Labs' GitHub organization.
[INDIRECT_PROMPT_INJECTION]: The skill consumes untrusted data from web pages, which is identified as a potential injection surface.
Ingestion points: Data is retrieved via snapshot, get text, read, and network requests commands.
Boundary markers: The documentation explicitly includes a 'Trust boundaries' section advising the agent to treat all page content as untrusted data and ignore embedded instructions.
Capability inventory: The skill uses subprocess calls to the agent-browser CLI, performs file writes for screenshots and session state, and conducts network operations through the browser.
Sanitization: Instructions require the agent to flag suspected injection attempts to the user rather than acting on them autonomously.

agent-browser