Skip to content
Get packs

Diagnose stuck TPU slice MIGs

Walks an agent through read-only diagnosis of GKE TPU slice failures caused by unstable Compute Engine MIGs.

Source Download skill
Repository
google/skills
Stars
21,782
Licence
Apache-2.0
Files
2 files

What it does

When a Google Kubernetes Engine (GKE) cluster fails to create a TPU slice (a group of Google AI accelerator chips used for training or serving models) or a JobSet stays pending, this skill tells the agent how to investigate. It checks the slice status first, then whether the underlying Compute Engine instance groups are stable, and reports likely causes. The agent is instructed to run only read-only commands and to give fixes to the user rather than apply them.

Great for

  • Platform or ML engineers running TPU workloads on GKE who see SliceCreationFailed errors
  • Cloud operators debugging JobSets stuck in pending because node pools are still updating

Skip it if

  • Teams whose slice failures are unrelated to MIG stability; the skill itself points them to the separate dynamic-slices monitoring skill
  • Anyone expecting automatic repair: the skill is explicitly read-only and will not drain, delete, or recreate nodes

Try asking

  • “My TPU slice creation failed with SliceCreationFailed and the JobSet is pending. Help me find out if the MIG is unstable.”
  • “Check whether the managed instance group behind my GKE TPU node pool reports isStable=False.”
  • “Why is my JobSet stuck in pending on cluster my-cluster in us-central1? Investigate the MIG status.”
Step-by-step guide: how to use Diagnose stuck TPU slice MIGsInstall, first run, examples, tips and FAQs

Install in 3 steps

  1. 1

    Download the skill

    The ZIP contains the 2 original files, the Apache-2.0 licence and an attribution note.

    gke-ai-troubleshooting-tpu-slice-mig-stuck-updating.zip
  2. 2

    Add it to your agent

    Claude Code (all projects)

    mkdir -p ~/.claude/skills && unzip gke-ai-troubleshooting-tpu-slice-mig-stuck-updating.zip -d ~/.claude/skills/

    Claude Code / Agent SDK (this project only)

    mkdir -p .claude/skills && unzip gke-ai-troubleshooting-tpu-slice-mig-stuck-updating.zip -d .claude/skills/

    Claude apps (web/desktop): Settings → Capabilities → Skills → Upload skill, and choose the ZIP.

  3. 3

    Ask for the task

    No special command needed — the agent reads the skill's description and loads it when your request matches. Try one of the prompts above.

Get better results

  • Give the agent your project ID, cluster name, location, node pool name, and approximate time of failure up front so it can scope the investigation window.
  • Make sure your gcloud identity has read-only viewer roles; the skill only needs those to inspect MIGs, logs, and Slice resources.
  • Review any fix the agent suggests before running it yourself, since the skill deliberately leaves changes to you.

What's inside

  • references/attribution.md
  • SKILL.md

Redistributed unmodified from google/skills under the Apache-2.0 licence. The assessment was drafted with AI from the skill's files on 2026-10-29. Skills can instruct your agent to run code — read SKILL.md and any scripts before installing.