The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The secretGenerator example in SKILL.md includes a hardcoded secret value (api-key=secret123). Hardcoding secrets in configuration files is an unsafe practice that can lead to credential leakage if adopted by users or agents.
[EXTERNAL_DOWNLOADS]: The skill provides examples for fetching Kubernetes resources from remote, non-standardized sources such as github.com/org/manifests and raw.githubusercontent.com/org/repo. Fetches configuration from the well-known Prometheus Community Helm repository. Downloading remote manifests introduces a risk of deploying malicious or unvetted infrastructure if the remote source is compromised.
[COMMAND_EXECUTION]: The documentation includes commands that pipe the output of kustomize build directly into kubectl apply. This pattern encourages the deployment of resources without a manual review step, which is particularly risky when combined with remote resource fetching.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection via external data ingestion.
Ingestion points: Remote Git repositories and URLs specified in the resources and helmCharts fields of the kustomization files.
Boundary markers: None present. There are no instructions for the agent to validate the integrity or content of the remote files before processing them.
Capability inventory: The skill uses kubectl apply, which grants the ability to create, modify, or delete any resource within the Kubernetes cluster, providing a high-impact capability if the ingested data contains malicious instructions.
Sanitization: None present. The content is used directly in manifest generation and deployment.

kustomize