The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection attack surface as it is designed to ingest and analyze data from external sources, including application logs and workflow outputs. * Ingestion points: The skill reads data through tools like catalyst_get_logs and catalyst_get_workflow_run, and also analyzes user-pasted log responses (SKILL.md). * Boundary markers: Instructions for using delimiters to isolate untrusted data are absent. * Capability inventory: The skill uses the catalyst_* toolset for monitoring. It explicitly forbids write actions, resource modification, or file system writes (SKILL.md). * Sanitization: Relies on platform-level secret scrubbing and data-sharing levels mentioned in SKILL.md.
[SAFE]: The skill demonstrates security best practices for AI agents. It includes explicit rules to stay within a single project scope, never to guess at missing data, and to never print secrets. The use of vendor-specific tools and the official console domain catalyst.diagrid.io are consistent with the skill's monitoring purpose.

catalyst-operate