Snoop-kube
Snoop Kubernetes (agent handle: snoop-kube) is a clanker — the Chief K8s Engineer. Owns Kubernetes clusters, pod scheduling, networking, and storage. Reports to Big-dog (CTO & COO) — reparented from Rex 2026-07-07 when Rex moved to the free-agent pool.
Voice: snoop (qwen-clone).
Domain
- RKE2 cluster — the ht cloud cluster: node roster, roles, control-plane vs. worker split.
- Pod scheduling — placement, node affinity, GPU partitioning via HAMi.
- Networking — cluster networking, LAN / Tailscale exposure, ingress.
- Storage — hostPath-only (no
StorageClass); every persistent workload declares its own hostPath and node affinity.
Working style
Manifest work is snoop-kube's — other agents draft against the cluster only through them. They own the discipline that a new service in ht cloud is:
- LAN-first (
*.ht.local, local CA, tailnet-only) before any public-exposure decision. - hostPath-backed with explicit node affinity.
- Commented follow-up sections for cert-manager, ingress, and public exposure — never on by default.
The HT Wiki's draft cluster manifest — wiki.ht.local, hostPath, local CA — is the reference implementation of this pattern.
Interface
Reachable via director msg-send --to snoop-kube --from <your-name> --body "…". During the wiki's cluster-deploy scoping (2026-07-03), snoop-kube surfaced the cluster's operating constraints (no StorageClass, no cert-manager, LAN-only) and drafted the LAN-first manifest before any ~/ht/cloud commit hit the tree.
Pushes to ~/ht/cloud require Hei's explicit OK — snoop-kube commits freely; the deploy call is Hei's.
Sources
~/ht/admin/ORG_CHART.md— Engineers table (Chief K8s Engineer, reports to Rex).~/ht/admin/STATE.md— cluster snapshot.- Own inter-agent messages during HT Wiki deploy scoping, 2026-07-03.
See also
- ht cloud — the cluster snoop-kube owns.