garmin-insights

name: garmin-insights description: Generates structured Garmin coaching insights (daily, weekly, post-run) from derived metrics and stores them via the AI insight pipeline. metadata: short-description: Garmin AI insight pipeline orchestration skill.

When to use

• User asks for Garmin-derived coaching insights.
• A sync/derive flow needs to trigger insight generation.
• You need a concise AI summary from Garmin metrics without exposing raw JSON.

Workflow

1. Read normalized metrics from:
   • training_derived_metrics
   • garmin_daily_metrics
   • garmin_activities
2. Build a structured summary using v0/lib/garminInsightBuilder.ts.
3. Generate insight text through:
   • v0/app/api/ai/garmin-insights/route.ts (streaming API), or
   • v0/lib/server/ai-insights-worker.ts (background job).
4. Persist into ai_insights with evidence metadata.

Insight types

• daily
• weekly
• post_run

Guardrails

• Never send raw Garmin JSON to the model.
• Keep context under 800 tokens.
• Include a safety disclaimer in the system prompt.
• Emit telemetry events:
  • ai_insight_created
  • ai_safety_flag_raised

Integration points

• Queue payload + enqueue: v0/lib/server/garmin-sync-queue.ts
• Worker: v0/lib/server/ai-insights-worker.ts
• Shared generation service: v0/lib/server/garmin-insights-service.ts