The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of the kubectl command-line tool to interact with the Kubernetes cluster.\n
Evidence: The SKILL.md file and scripts/diagnose.sh script execute numerous commands including kubectl get, kubectl describe, kubectl top, kubectl logs, and kubectl exec.\n
Context: Specifically, kubectl exec is used to run diagnostic tools like redis-cli, mount, and du inside active containers in the crs namespace.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its processing of untrusted data from the cluster environment.\n
Ingestion points: The agent ingests data from container logs via kubectl logs and from the output of commands run inside containers via kubectl exec (found in SKILL.md and scripts/diagnose.sh).\n
Boundary markers: No delimiters or explicit instructions are provided to the agent to distinguish between legitimate diagnostic output and potential instructions embedded within that output.\n
Capability inventory: Across the provided files, the skill demonstrates the ability to execute code in containers, read application logs, and access cluster events.\n
Sanitization: There is no evidence of logic designed to sanitize or validate the content retrieved from logs or container outputs before it is presented to the agent.

debug-buttercup