The Agent Skills Directory

Sanitized Data Ingestion: The skill utilizes a dedicated JQ script (scripts/sanitize-readiness-input.jq) to process Kubernetes manifest data before it is evaluated. This script acts as a security boundary by projecting only specific, non-sensitive fields (like resource requests, security context metadata, and image policies) while excluding potentially sensitive data fields from Secrets or ConfigMaps. This approach follows the principle of least privilege and data minimization.
Read-Only Assessment Logic: The instructions explicitly define the assessment as read-only and strictly prohibit commands that modify the cluster state (such as kubectl apply or az aks update). This constraint ensures the skill does not inadvertently alter production environments during the readiness check.
User-Controlled Remediation: Every proposed fix for identified incompatibilities is presented as a diff, and the skill is instructed to wait for explicit user approval before applying any changes to local files. This manual checkpoint ensures that the user retains full control over their configuration files.
Indirect Prompt Injection Mitigation: Although the skill processes untrusted Kubernetes manifests, the combination of a strict YAML constraint specification and the allowlist-based JQ projection provides a strong defense against malicious instructions embedded in workload data.

aks-automatic-readiness