Debug a running LlamaCloud release

Goal: take "something is broken" and produce a structured diagnosis: which pod, what symptom, which values.yaml field (or external dependency) is the root cause, and what to change. Read-only against the cluster.

For known error signatures, the catalog is in patterns.md next to this file. Read it lazily — only when you've narrowed down to a candidate symptom — to keep this skill's context budget small.

Step 0 — Confirm context

kubectl config current-context
helm list -A | grep -i llama

Read the context back to the user and confirm. Identify the release(s). If multiple, ask which one.

Step 1 — Health snapshot