The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied application requirements (e.g., 'Deploy: Node.js API...') to generate Kubernetes manifests and Helm charts. This creates an attack surface where malicious instructions could be embedded in the data provided by the user.
Ingestion points: User input describing infrastructure stacks, deployment requirements, and tenant configurations as seen in the Examples section of SKILL.md.
Boundary markers: Absent. There are no explicit delimiters or system instructions defined to prevent the agent from obeying instructions embedded within the user-provided data.
Capability inventory: The skill documentation includes high-privilege commands such as kubectl exec for container shell access, helm upgrade --install for cluster modifications, and kubectl logs for data retrieval.
Sanitization: Not defined within the skill logic. While the guidelines recommend external image scanning with Trivy, there is no validation of the generated YAML or shell commands against injection attacks.

kubernetes-helm