/ratel-langfuse-dashboards — design Langfuse dashboards for an agent

Pair this skill with /ratel-langfuse-instrument. That skill writes the trace vocabulary; this one builds the dashboard specs that turn that vocabulary into visible value — both Ratel's value and general agent-health visibility.

The output is a markdown plan with one section per dashboard. The customer builds the dashboards by clicking through the Langfuse UI; this is intentional. Auto-creating dashboards via API requires assumptions about the Langfuse instance that we don't make in v1.

Why two groups of dashboards

Partner startups want two different stories from the same data:

1. "Ratel is moving the numbers" — token spend down, retrieval quality up, fewer "tool not found" errors, lower cost per session. These dashboards justify the engagement to whoever signs the cheque. The set evolves as Ratel ships new features (today it's pre-filter + retrieval; v0.1.7 adds skills; v0.1.9 adds suggestion adoption; v0.1.12+ adds semantic + re-rank). The shipped ones go in the customer's dashboard plan; the roadmapped ones get a "we'll add this when X ships" footnote.

2. "Our agent is healthy" — latency percentiles, error rates per tool, abandoned-session rates, score distributions. These are useful regardless of Ratel. They build trust because they help the customer's engineers find their own bugs.

We always include both. Ratel-only dashboards feel like a sales pitch; agent-health-only dashboards feel like we forgot why we're there.

Workflow

Step 1 — Read prior instrumentation context