The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documentation includes curl command examples intended for the agent to execute in order to send data to the Ray application's local server at http://localhost:23517/. While these commands target the local machine, they involve shell-level execution.
[INDIRECT_PROMPT_INJECTION]: The skill allows the rendering of arbitrary HTML content through the custom payload type. This creates a surface for indirect injection if the agent processes untrusted data and includes it in the HTML payload without proper sanitization.
Ingestion points: Data ingested from external sources (e.g., file content, web page summaries) that is subsequently interpolated into the content field of a custom payload in SKILL.md.
Boundary markers: Absent. The skill does not define specific delimiters or instructions to ignore embedded code within the HTML content.
Capability inventory: The skill uses POST requests to a local HTTP server and provides curl command templates.
Sanitization: Absent. There are no instructions for the agent to escape, filter, or validate the HTML before sending it to the Ray application.

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