health-drop-diagnosis

name	health-drop-diagnosis
description	Explains a health score drop — decomposes which score components moved, finds the triggering events, and recommends a proportionate response. Use when a health score fell, a risk alert fired, or someone asks "why did X's health drop".
license	MIT
compatibility	Tool calls resolve natively in Quivly, or in Claude Code / claude.ai / any MCP-enabled agent via the Quivly MCP connector. Without tools, the skill degrades to a guided manual workflow.
metadata	{"author":"quivly","version":"1.0","category":"customer-engineering","quivly-tools":"health.get-health-score health.get-insights platform-data.get-usage platform-data.search-tickets communication.search-conversations"}

Health Drop Diagnosis

You explain a health score movement in plain language: which inputs moved, what real-world events drove them, and how worried to be.

Core principle: a health score is a smoke detector, not a diagnosis — never recommend action from the score alone; trace it to real events first.

Workflow

1. Get the current health score with its component breakdown (get-health-score). Identify which components dropped and by how much — the composite hides the story.
2. For each falling component, find the underlying event:
   • Usage component → pull usage detail (get-usage)
   • Support component → recent tickets (search-tickets)
   • Engagement component → conversation recency and tone (search-conversations)
3. Check signals on record (get-insights) for events the score may lag: champion departure, funding news, org changes.
4. Judge severity honestly:
   • Mechanical — one input crossed a threshold, no real-world change (e.g., a holiday week tanked logins)
   • Early warning — real deterioration, early stage, recoverable
   • Confirmed risk — multiple components falling together with corroborating events

Output Format

Health drop: {Customer} — score before → after, severity verdict

What moved — component-by-component, each mapped to its real-world cause

The story — 2-3 sentences of narrative: what is actually happening at this account

Response — proportionate to severity: nothing (mechanical), targeted play (early warning), or escalate to save-plan mode (confirmed)

Guidelines

• A score is a smoke detector, not a diagnosis — always trace to real events before recommending action.
• Call out mechanical drops confidently; false alarms erode trust in the whole scoring system.
• If severity is "confirmed risk", recommend running a churn save plan rather than duplicating one here.

Related skills: usage component is the driver → usage-drop-investigation; confirmed risk → churn-save-plan; full structured review → customer-health-review.