Dashboards and real-time visualization skill for Codex.
The dashboards and real-time visualization skill makes Codex treat a live dashboard as a system, not a screenshot. It asks about update cadence, latency and render budgets before picking charts. It ships in OpenAI’s Build Web Data Visualization plugin.
Updated November 12, 2026 · 3 min read
At a glance
| Skill name | dashboards-and-real-time-visualization |
| Comes with | The Build Web Data Visualization plugin, by OpenAI |
| Install | codex plugin add build-web-data-visualization@openai-curated, or /plugins in Codex |
| Codex uses it | When you ask for monitoring views, streaming charts, linked views, downsampling or a fast operational dashboard |
| Source | openai/plugins on GitHub |
What it does
The skill starts with questions: is the data an append-only stream, polling, bursts or full snapshots? What latency matters? What must the user notice or do? How many charts are on screen at once? Then it applies rules in four areas:
- Real-time: clear time windows, live data kept apart from history, explicit handling of late and missing events, and the last good chart kept visible during reconnects with distinct stale, offline and error states.
- Layout: current state first, then context, then controls. Filters sit next to the views they change, and not every tile gets equal weight.
- Interaction: overview first, then filter, then detail on demand; brushing and linking; selections that survive updates; a clear reset path.
- Performance: cut marks and aggregate before optimizing code, pick SVG, Canvas2D or WebGL on purpose, and use ring buffers and viewport culling for long-running views.
Mobile is the default. The skill plans a portrait view, touch input, spotty connections and alerts unless you rule mobile out. Its output names the update model, failure modes and render budgets before any widgets get built.
When to use it
Install it when you’re building an operations or monitoring screen fed by a WebSocket, polling or a stream. It stops the usual problems with AI-built dashboards: a grid of equal tiles, charts that freeze or jump on reconnect, and pages that slow down as data piles up.
Try it
Use the dashboards and real-time visualization skill. Build an operations dashboard from our WebSocket feed of order events: orders per minute, error rate with an alert threshold, and a latency chart. Show last-updated time, handle reconnects and stale data, and make the mobile view work first.
Does it work in Claude Code?
The instructions are a standard SKILL.md, so you can copy the folder from GitHub into ~/.claude/skills. It links to shared theory files in the plugin’s top-level references folder and to the sibling testing skill, so copy those too or Claude Code will only have the main file.