The Agent Skills Directory

[NO_CODE]: The skill contains no executable scripts or binary files, consisting solely of markdown instructions and metadata configuration.\n- [EXTERNAL_DOWNLOADS]: Provides instructions for the agent to download and install the agent-browser utility from official npm registries and trusted GitHub repositories.\n- [COMMAND_EXECUTION]: Enables the agent to perform web automation by executing CLI commands for browsing, capturing snapshots, and interacting with page elements.\n- [REMOTE_CODE_EXECUTION]: Includes an eval command that allows the execution of arbitrary JavaScript within the target website's browser context, which is a standard feature for automation workflows.\n- [PROMPT_INJECTION]: As a browser-based agent, it is exposed to indirect prompt injection where content from visited websites could attempt to influence the agent's behavior.\n
Ingestion points: External website content retrieved via snapshot, get, and find commands.\n
Boundary markers: The skill documentation does not define specific delimiters for separating web content from agent instructions.\n
Capability inventory: Extensive interaction capabilities including form filling, JavaScript execution, and file uploads.\n
Sanitization: Relies on standard browser environment isolation; no specific content sanitization is described before data is returned to the agent.

Agent Browser