The Agent Skills Directory

[Indirect Prompt Injection] (HIGH): The skill ingests untrusted data from Kubernetes pod logs, events, and descriptions. Malicious actors could place instructions in these fields to take control of the agent session. Ingestion points: get_pod_logs (SKILL.md, diagnose-pod.py), get_events (SKILL.md, diagnose-pod.py), and describe_pod (SKILL.md, diagnose-pod.py). Boundary markers: Absent. No instructions or delimiters distinguish diagnostic data from system instructions. Capability inventory: High-privilege tools including kubectl_exec (arbitrary command execution in containers) and cluster-wide resource access. Sanitization: Absent. External content is processed directly.
[Command Execution] (HIGH): The skill provides the agent with kubectl_exec capability (SKILL.md), allowing for arbitrary code execution within the managed Kubernetes cluster. While intended for troubleshooting, it represents a significant security surface if the agent is misled by indirect injection.
[Best Practice Violation] (LOW): The scripts/health-check.sh script does not properly quote variables when constructing commands (e.g., kubectl $KUBECTL_OPTS), which could lead to unexpected behavior if inputs contain spaces or special characters.

k8s-troubleshoot