The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the belt CLI tool to perform all browser operations. The provided templates and documentation also reference infsh, which serves as a functional alias for the belt command provided by the vendor. These tools are used for legitimate browser session management and interaction.\n- [PROMPT_INJECTION]: The skill has an inherent attack surface for indirect prompt injection because it processes untrusted data from the web without explicit boundary markers.\n
Ingestion points: External data enters the agent's context through browser functions such as open, snapshot, and execute (which returns page content or JavaScript results).\n
Boundary markers: None are present; the skill instructions do not provide specific markers to help the agent differentiate between its own directives and the content of the web pages it is browsing.\n
Capability inventory: The skill provides high-privilege capabilities within the browser context, including arbitrary JavaScript execution (execute), navigation (open/goto), and comprehensive element interaction (interact).\n
Sanitization: No content sanitization or instruction-filtering is performed on the data retrieved from the browser before it is presented to the agent.

agent-browser