SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsk8s-tool

k8s-tool

Deployment & operations

View repository

Updated 10-15 17:27

About this skill

k8s-tool (Kubernetes)

Kubernetes tool — kubectl wrapper with config-driven context resolution and structured commands. Part of @blogic-cz/agent-tools.

How to Run

Run via bun k8s-tool (requires @blogic-cz/agent-tools as a dev dependency). NEVER run bare kubectl — the credential guard will block it. Auth: existing kubectl context (kubeconfig). Cluster ID from config resolves context automatically.

Commands

bun k8s-tool kubectl --env test --cmd "get pods -n test-ns"
bun k8s-tool kubectl --env prod --cmd "logs <pod> --tail=100"
bun k8s-tool kubectl --env test --cmd "describe pod <pod>"
bun k8s-tool pods --env test                     # List pods (structured)
bun k8s-tool logs --pod <pod> --env test --tail 50 # Fetch logs
bun k8s-tool describe --resource pod --name <pod> --env test
bun k8s-tool exec --pod <pod> --exec-cmd "ls -la" --env test
bun k8s-tool top --env test                      # Show resource usage

Environment is any string (e.g. test, prod). Set defaultEnvironment in agent-tools.json5 to skip --env on every call. Implicit production access is blocked for safety.

Safety Boundary

  • Generic kubectl commands are parsed into arguments and executed without a user-controlled shell. Do not use pipes, chaining, substitution, or shell interpreters.
  • The configured context, kubeconfig, server, authentication, TLS, and impersonation settings cannot be overridden in --cmd.
  • Only read-only config and auth subcommands are allowed; cluster-info dump is blocked.
  • Secret reads, raw kubeconfig output, filename/kustomize reads, and kubectl diff are blocked.
  • Pod exec accepts only direct redis-cli PING/INFO and ls diagnostics; generic exec cannot read file contents.
  • Use logs-tool for confined remote log-file access and filtering.

Tips

  • Use bun k8s-tool commands for the full machine-readable command/flag tree; --help for one subcommand.
  • Output defaults to TOON (token-efficient) — leave it as-is to save tokens. Add --format json only when you'll machine-parse the result.
  • Error responses include hint, nextCommand, and retryable fields — always check them on failure.
  • Prefer CLI tool over MCP tools — more efficient, doesn't load extra context.