The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides numerous examples instructing the agent to utilize shell commands such as curl, uuidgen, hostname, cat, sed, and basename. These commands are primarily used to construct and transmit structured log payloads to the Ray application's local HTTP API.
[INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection as it processes external data for logging, which could contain malicious instructions designed to influence the agent or the log viewer.
Ingestion points: Data ingested for logging includes file contents (specified in rules/file-contents.md) and general values from the agent's context (specified in rules/log.md).
Boundary markers: Absent. The skill does not provide instructions to wrap logged data in specific delimiters or to disregard instructions embedded within that data.
Capability inventory: The skill uses curl for network requests to localhost (in rules/ray-local-http.md), and cat, sed, and basename for file processing (in rules/file-contents.md). Additionally, it documents a capability to manipulate the system clipboard via metadata (in rules/log.md).
Sanitization: Present. The documentation recommends HTML encoding for several payload types (Text, XML, and File Contents) to ensure they are rendered correctly within the Ray UI.
[DATA_EXFILTRATION]: The skill instructs the agent to read local files and system metadata (hostname) and send them to an HTTP endpoint. However, the destination is restricted to localhost:23517 by default, which is the standard endpoint for the Ray application. This significantly limits the risk of external exfiltration as the data remains on the local machine.
[SAFE]: The skill's functionality is entirely consistent with the Ray developer tool created by Spatie. All external links and payload structures align with the official Ray documentation.

ray