| name | talos-inspect |
| description | Read-only Talos OS triage for Anton. Use to check cluster health, node readiness, etcd quorum, Talos services, disks, NVMe inventory, install disks, routes, interfaces, MTU, VIP, inter-node reachability, DNS, or suspected node-level failures. |
Talos Inspect
Goal: Diagnose Anton below the Kubernetes workload layer with read-only Talos and kubectl commands.
Success means:
- Health, disks, and network checks use the repo talosconfig.
- The report names the first failing layer and the evidence behind it.
- Findings hand off mutating fixes to an operator-approved workflow.
Stop when: the Talos layer is classified as healthy, degraded, or blocked by connectivity/tooling.
Workflow
- Read
talos/AGENTS.md.
- Pick the relevant reference:
references/health.md for node, etcd, service, and control-plane health.
references/disks.md for install disks, NVMe, volumes, and Longhorn disk inputs.
references/network.md for interfaces, routes, bonds, VIP, DNS, MTU, and reachability.
- Run read-only checks only.
- Return a ranked finding list with command evidence and the next safe handoff.
Command Shape
mise exec -- task talos:health
The wrapper uses the current Tailscale endpoint map, probes each node
individually, and then runs the server-side health check. For a direct query,
consume the same resolver instead of copying endpoints into the skill:
NODES="$(python3 scripts/cluster-targets.py resolve --format addresses --show-addresses)"
ENDPOINT="${NODES%%,*}"
mise exec -- talosctl --talosconfig ./talos/clusterconfig/talosconfig \
--endpoints "$ENDPOINT" --nodes "$NODES" <read-only-command>
Do not use the LAN addresses from the generated talosconfig off-LAN.
Boundaries
Treat apply-config, upgrade, reset, bootstrap, reboot, and etcd member changes as mutation handoffs. Provide the command and safety gates; wait for the operator to approve execution.