Agent Skills Dashboard
Sample observability UI for agent runs — KPI strip, model latency-vs-success scatter, and per-tool accuracy. Metrics are placeholders today; the page is the aggregation surface that would sit over a production telemetry table.
Sample observability UI — metrics are placeholders. The dashboard renders from a static TypeScript mock dataset, not a live telemetry pipeline. The intent is to show the observability surface — what aggregations matter and how they’re laid out — not to claim live production numbers.
A high-signal technical showcase of what an observability layer for the other portfolio agents would look like, rendered against a representative mock dataset.
What’s on the page
KPI strip — total runs, weighted success rate, weighted average latency, and tool accuracy across all models in the dataset.
Success Rate vs. Latency (scatter) — each dot is a model, sized by total runs. Reveals the latency-vs-quality frontier: cloud models cluster top-right (slower, more accurate); local Llama variants trade quality for speed and cost.
Tool Usage Accuracy (bar) — per-tool success rate across all agent invocations in the dataset, sorted descending.
Model breakdown table — model · hosting (cloud/local) · success · latency · runs · cost per 1k runs, with horizontal scroll on phones.
Live playground
The ReAct tool-use playground used to live at the bottom of this page; it now has its own route at /playground. Submit a query and watch the agent’s thoughts, tool calls, and observations stream in over SSE.
Data Source
Today: a static module at src/lib/agent/dashboard-data.ts exports MODELS, TOOLS, and a summarize() aggregation, with numbers chosen to look plausible for the model families involved.
Planned: every agent run from the other showcases writes a row to a Postgres agent_runs table — model, tool calls, latency, success, cost — and this page becomes a thin read-only aggregation layer over it.