The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill acts as a bridge to send data to the Ray desktop application, which supports rendering raw HTML payloads and displaying desktop notifications. While this is a core feature for debugging, it creates a surface where untrusted data processed by the agent could potentially be rendered in the host application.
Ingestion points: User-provided content or application logs passed to the agent to be 'sent to Ray'.
Boundary markers: The skill defines a structured JSON schema but does not include specific instructions for the agent to sanitize or escape data before transmission.
Capability inventory: The skill utilizes network POST requests to localhost:23517 to transmit data.
Sanitization: The skill does not implement internal sanitization; protection against potentially malicious HTML or scripts relies on the Ray desktop application's internal security guardrails.
[DATA_EXPOSURE]: The skill instructions prompt the inclusion of an origin object containing local file paths and hostnames. Because the target destination is restricted to localhost, this is considered a standard debugging practice and does not constitute a remote exfiltration risk.

debugging-output-and-previewing-html-using-ray