The Agent Skills Directory

Indirect Prompt Injection (LOW): The skill is susceptible to instructions embedded in the external websites it processes.
Ingestion points: Data from untrusted external domains is ingested through browser_snapshot, browser_take_screenshot, and WebSearch results.
Boundary markers: There are no delimiters or explicit instructions provided to the agent to disregard instructions found within the external content.
Capability inventory: The skill allows the agent to interact with pages via browser_click and browser_type, which could be exploited if a malicious site contains instructions the agent follows.
Sanitization: No validation or sanitization of the external web content is performed before the agent processes or interacts with it.

competitor-scan