The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to how it handles data from the cluster environment.
Ingestion points: The workflow involves reading untrusted content via kubectl logs and kubectl get events as specified in SKILL.md.
Boundary markers: No delimiters or explicit instructions are provided to the agent to treat the ingested log/event content as data rather than instructions.
Capability inventory: The skill utilizes kubectl and flux CLI tools for resource investigation.
Sanitization: There is no mechanism described to sanitize, truncate, or filter potentially malicious instructions embedded in pod logs or event messages before they are processed by the agent.

failing-deployments