The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/check_namespace.py utilizes subprocess.run to execute kubectl commands. While it uses argument lists rather than shell strings to mitigate direct shell injection, it represents an active system command execution surface driven by user-supplied namespace arguments.
[COMMAND_EXECUTION]: The documentation provides numerous examples of high-privilege operations, including ssh access to nodes for running systemctl, journalctl, and disk benchmarking via dd (found in references/performance_troubleshooting.md). While these are standard tools for troubleshooting, they provide a powerful execution path that could be misused if the agent is manipulated.
[EXTERNAL_DOWNLOADS]: The references/helm_troubleshooting.md file recommends installing Helm plugins from external GitHub repositories (databus23/helm-diff and jkroepke/helm-secrets). These are recognized community tools and are documented neutrally as part of the troubleshooting workflow.
[INDIRECT_PROMPT_INJECTION]: The diagnostic logic processes untrusted data from the Kubernetes cluster environment, which could be exploited to influence the agent's behavior through malicious metadata.
Ingestion points: The scripts/check_namespace.py script parses JSON output from kubectl get commands for pod statuses, events, and resource metadata.
Boundary markers: Absent. There are no explicit instructions or delimiters for the agent to treat data returned by the diagnostic scripts as untrusted or potentially adversarial.
Capability inventory: The skill provides extensive cluster access capabilities, including kubectl exec, kubectl apply, and node-level ssh access.
Sanitization: Absent. The script passes raw strings retrieved from the Kubernetes API (such as pod names and event reasons) directly into the output strings presented to the agent.

k8s-troubleshooter