helm-charts

Diagnose problems with an installed LlamaCloud release — pods crashing, services not reachable, login broken, parses failing, OCR queue stuck, etc. Use when the user says "something's wrong", "pods are stuck", "X is failing", "the install came up but something is broken", or pastes a kubectl error. Performs a structured symptom-to-cause walk using read-only kubectl and cloud CLI commands, then matches against the LlamaCloud-specific pattern library in `patterns.md`. Produces a `debug-report.md` safe to share with LlamaIndex support.

Run `helm install` (or `helm upgrade --install`) for the LlamaCloud chart, monitor the rollout in real time, surface failing pods immediately, and capture a diagnostic bundle if anything fails. Use when the user says "install the chart", "run helm install", "deploy llamacloud", "upgrade", or naturally follows a successful preinstall-check. Runs with `--wait --timeout` and intentionally **does not** pass `--atomic`, so failed pods remain in place for the `debug` skill to investigate. Read-only against the cluster except for the helm install itself.

Validate a values.yaml and verify cluster + dependency reachability before running `helm install` for the LlamaCloud chart. Use when the user says "preinstall check", "validate my values", "check before install", "preflight", or before any first-time install or upgrade. Performs static schema validation, cluster capacity inspection, local network probes, and read-only routing checks against managed cloud services (AWS / Azure / GCP). Fully read-only by default; only one optional phase can create a short-lived Job, and only with explicit user consent.