Analyze cluster health
Turn a cluster health snapshot into a verdict, degraded components, and the next checks worth running. Use for health questions and pre/post-change checks.From its SKILL.md
npx -y skills add qa-veritas/skillpack --skill analyze_cluster_healthAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
1.5 KB, 293 tokens by cl100k_base, as published. Nobody here has run it
analyze_cluster_health
Purpose
A health endpoint returns a wall of JSON. Convert it into a single verdict (healthy / degraded / down), name the components that are actually broken, and point at the next check — don't recite every field.
Inputs
health_snapshot_json— overall status, per-node status, replica / shard / quorum counts, pending tasks.node_list(optional) — expected membership, to catch a node that silently dropped out.
Outputs
overall_verdict, quorum status, degraded_components (with the
specific issue and how long), next_checks, and a confidence level.
Prompt template
State the verdict in one word and justify it. Check quorum / control-plane availability first — a degraded data plane with intact quorum is recoverable; lost quorum is an emergency. List only genuinely degraded components. A node missing from the snapshot entirely is worse than one reporting unhealthy. Recommend the next 1-3 localizing checks. Diagnose only; do not remediate here.
Examples
- One node NotReady, replicas unassigned, quorum intact → degraded, not down; next check is the node, then allocation.
- Two of three control-plane nodes unreachable → down, quorum lost; escalate.
What ships with it
Read from the repository
Just SKILL.md. No reference files, no scripts.