How to use the Visualize Data agent skill: setup and examples

What the Visualize Data skill makes the agent do

The skill changes how the agent approaches a chart. Instead of drawing the first plot that fits the request, the agent first states the analytical question the chart should answer and the takeaway it should support. It then checks the data: grain, measures, dimensions, units, missing values, filters, time window, sample size, and source.

Chart type is chosen from the relationship in the data. The skill maps relationships to chart families: a line or area chart for change over ordered time, a bar chart for category comparison, a histogram, box plot, or density plot for distributions, a scatter plot for relationships between numeric measures, stacked bars or areas for composition, a funnel or ordered bars for stage progression, a waterfall for contributions to change, and a table for exact lookup values. A pie chart is used only for a small set of meaningful parts.

• States the question and intended takeaway before drawing
• Verifies the data before encoding it
• Picks a chart family from the relationship in the data
• Labels axes and units, sorts and scales sensibly, and uses restrained color
• Recomputes plotted values from the reviewed rows as a QA step

What is an agent skill?

An agent skill is a folder of instructions that an AI agent can load when a task matches it. The main instructions live in a file named SKILL.md. This skill's folder also contains agents/openai.yaml. Once the skill is installed, you do not run it yourself; you ask the agent for a chart and the agent follows the skill's workflow.

Install the Visualize Data skill

Download the skill ZIP from this page and unzip it. Then place the resulting visualize-data folder in one of these locations, depending on how you want to use it:

• Personal skills: ~/.claude/skills/
• Project skills: the .claude/skills/ folder inside a specific project
• Claude apps: open Settings, go to Capabilities, then Skills, and upload the skill there

Trigger the skill with natural requests

You do not need a special command. Describe the chart you need and what it is for. The skill's description covers designing, creating, revising, or QA-ing charts for reports, dashboards, notebooks, slides, files, or inline analytical answers, so requests that mention any of these tasks are the ones it is meant for. Naming the audience and the destination helps the agent pick the right output format.

• "Make a line chart of monthly signups for the board slide, and check the values against this CSV."
• "Review this bar chart for misleading axes before I put it in the report."
• "Create a waterfall showing what drove the change in revenue between Q1 and Q2."
• "Draw a reproducible Matplotlib figure of these survey results for the appendix."

Examples of what to expect

When you ask for a chart, expect the agent to ask or state the question, confirm the data's grain and units, choose a chart family, and then render. Rendering uses the destination's native system when one is selected. Otherwise the skill prefers reproducible Python with Matplotlib or SVG for files and inline artifacts, and notebook-native plotting for notebooks.

For a revision or QA request, the agent should recompute plotted values from the source rows, check ordering, scales, labels, clipping, and overlap, and confirm the exported file opens and contains the expected marks and text. The skill also asks the agent to keep the code needed to reproduce the chart when you need an auditable artifact.

What you need before you start

The skill text itself does not require API keys or accounts. It does not run scripts and does not use the network. File and inline outputs imply a Python environment with Matplotlib or another plotting library, but the files do not pin a specific library or version, so you install whichever plotting tool your project uses.

• Python with Matplotlib or another plotting library for file and inline outputs
• Your source data, ideally with a clear grain, units, and documented filters
• A host that can display images or notebooks if you want the agent to inspect the final chart

Safety notes: scripts and network

According to the skill files, the skill does not run scripts and does not use the network. The skill also tells the agent not to install or require React, Recharts, Vite, MCP widgets, or proprietary UI renderers, and to avoid remote scripts unless you explicitly accept that dependency. For self-contained HTML, it steers toward inline SVG or canvas, or a static image with a semantic fallback. Because the agent may still write and run plotting code in your environment, review any code it produces before running it on sensitive data.

Tips for better charts

Give the agent the takeaway you want, not only the data. The skill is built around charts as evidence, so a clear question leads to a clearer chart. Ask for a source note instead of cramming provenance into the title, and ask for a data table when readers need exact values. If a chart is dense, ask for aggregation, faceting, or filtering rather than smaller text.

• Name the audience and destination (slide, report, notebook, file)
• Ask for a QA pass on an existing chart to catch truncated axes or wrong values
• Request the reproducible code if you need to audit the chart later
• Keep pie charts for a few meaningful parts only