The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted data from the GKE cluster environment.
Ingestion points: The skill reads live context from the cluster using kubectl get <resource> -o yaml, kubectl describe, and kubectl logs as defined in SKILL.md.
Boundary markers: There are no explicit markers or instructions provided to the agent to distinguish between its instructions and potentially malicious content embedded in the resource outputs or logs retrieved from the environment.
Capability inventory: The skill executes CLI commands (kubectl, gcloud) and generates new skills that are intended to execute similar commands.
Sanitization: The instructions do not specify any sanitization, filtering, or validation of the data retrieved from the cluster before it is used to generate the new skill draft.
[COMMAND_EXECUTION]: The skill relies on executing system commands for its core functionality.
Evidence: SKILL.md explicitly lists commands like kubectl get, kubectl describe, kubectl logs, and gcloud container clusters describe for investigating cluster state. The generated skill template in references/generated_skill_skeleton.md also centers around executing kubectl and gcloud commands for remediation.
[DATA_EXFILTRATION]: The skill accesses and processes potentially sensitive cluster information.
Evidence: The skill is designed to read Kubernetes logs, resource specifications (which may contain environment variables or configuration), and cluster metadata. While the instructions emphasize creating public-facing output and removing internal references, there is an inherent risk of sensitive data exposure if the agent or user does not properly review the generated content for secrets or private data.
[DYNAMIC_EXECUTION]: The skill dynamically generates new instructional content (skills) that perform system operations.
Evidence: The workflow in SKILL.md involves drafting a new SKILL.md file using a template (references/generated_skill_skeleton.md) based on runtime discovery. These generated skills are intended to be used for automated or semi-automated remediation of technical issues.

gke-skill-creator