The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes Kubernetes manifests and cluster configurations, which constitutes an attack surface for indirect prompt injection from potentially untrusted workload data.
Ingestion points: Workload data is ingested through kubectl command output or local workspace files during readiness assessments as described in azure-kubernetes-automatic-readiness/SKILL.md.
Boundary markers: The skill contains explicit instructions to ignore embedded instructions in data and provides clear boundaries for assessment scope.
Capability inventory: Capabilities include cluster management via az aks and resource manipulation via kubectl across multiple files.
Sanitization: The skill employs a mandatory JQ filter (scripts/sanitize-readiness-input.jq) that performs an allowlist projection, stripping sensitive fields like environment variables, secrets, and managed fields before the data reaches the AI agent.
[EXTERNAL_DOWNLOADS]: The skill references and suggests the retrieval of configuration files and tools from external repositories.
Evidence: The workflow in azure-kubernetes-app-deploy/phases/quick-deploy.md suggests installing the Gateway API CRDs from github.com/kubernetes-sigs/gateway-api and references container images from Microsoft Container Registry (MCR) and Docker Hub.
Context: These downloads target well-known services and trusted organizations, adhering to standard DevOps practices for AKS management.

azure-kubernetes