The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the Kubernetes cluster, creating an indirect prompt injection surface.
Ingestion points: The skill reads container logs using kubectl logs and cluster events using kubectl get events in scripts/subcommands/pod.sh and scripts/subcommands/events.sh.
Boundary markers: There are no explicit instructions or delimiters used to separate the ingested log and event data from the agent's instructions.
Capability inventory: The fix sub-command in scripts/subcommands/fix.sh allows the agent to modify the cluster state, including deleting pods, scaling deployments, and rolling back versions.
Sanitization: The skill mitigates this risk by employing a strict command whitelist and filtering input for shell meta-characters such as semicolons, pipes, and backticks.
[DYNAMIC_EXECUTION]: The fix sub-command in scripts/subcommands/fix.sh dynamically executes kubectl commands based on AI agent input. This behavior is restricted by a predefined whitelist of allowed verbs (rollout, scale, delete pod, cordon, uncordon) to prevent arbitrary command execution.
[COMMAND_EXECUTION]: The skill uses kubectl, ssh, and jq for cluster operations. It supports remote cluster management via SSH and implements security best practices by using printf %q for argument escaping to prevent command injection on the target system.

k8s-check-fix