The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill identifies and mitigates the risk of ingesting untrusted data from Kubernetes logs and metadata which could contain adversarial instructions.
Ingestion points: Command outputs from kubectl logs, kubectl get pods, and observe datastream view are processed in Phase 5 and Phase 2.
Boundary markers: The skill mandates the use of a wrap helper (detailed in references/untrusted-output.md) that utilizes per-read randomized nonces and XML-style tags to isolate external data.
Capability inventory: The skill executes administrative commands via kubectl, helm, eksctl, and the observe CLI.
Sanitization: The wrap helper includes a sanitizer using sed and tr to strip ANSI escape sequences and control characters, preventing terminal-based injection or visual manipulation of output.
[COMMAND_EXECUTION]: The skill uses administrative shell commands to orchestrate the deployment and verification of the Kubernetes explorer.
Evidence: Orchestrates commands like kubectl patch, helm install, and eksctl create fargateprofile within SKILL.md and references/fargate-reference.md to configure the cluster environment.
Scope: The commands are standard for Kubernetes resource management and are restricted to the vendor-specific tools and official Kubernetes utilities.

deploy-k8s-explorer