The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local system commands to process infrastructure manifests. Specifically, it uses helm template for Helm charts and kustomize build for Kustomize overlays. It also executes a local shell script, preview.sh, which launches a background Node.js process to serve a web-based migration plan preview.
[INDIRECT_PROMPT_INJECTION]: The skill processes arbitrary user-supplied infrastructure manifests that may contain adversarial instructions designed to influence the agent's migration strategy or inventory classification.
Ingestion points: Local directory paths containing Chart.yaml, kustomization.yaml, docker-compose.yml, or *.yaml files.
Boundary markers: The instructions lack explicit delimitation (e.g., XML tags or unique markers) for user-provided data during the analysis phase.
Capability inventory: The agent has the ability to execute shell commands (helm, kustomize, node), write files to the local workspace (.migrate-to-openchoreo/), and generate complex HTML/Markdown reports.
Sanitization: The skill relies on standard third-party renderers (helm, kustomize) to parse inputs; however, it does not implement additional sanitization of the resulting YAML before agent analysis.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill identifies and classifies sensitive Kubernetes resources such as Secrets and ConfigMaps. The migration plan instructs the user on how to seed these into external secret stores. The local preview server (server.cjs) includes path normalization and validation logic to prevent directory traversal attacks, ensuring it only serves files from the designated run directory.

migrate-to-openchoreo