Claude Code's /dataviz skill is a built-in skill, added in v2.1.198, that loads chart and dashboard design guidance into context: how to pick a chart type, how to lay out a data-heavy UI, and a runnable color-palette validator. You type /dataviz or just ask for a chart, and Claude follows the guidance instead of winging it.
That is the whole feature in a sentence. But there is a bigger point hiding inside it. /dataviz is a SKILL.md file, exactly like the ones you build yourself. Anthropic reached for the same extension mechanism you have. So this post does double duty: it teaches you the skill, and it uses the skill to teach you how good skills are built.
What is the /dataviz skill in Claude Code?
/dataviz is one of Claude Code's bundled skills. It arrived in the July 2026 changelog, where the entry reads, word for word: "Added /dataviz skill for chart and dashboard design guidance with a runnable color-palette validator." That landed in Claude Code v2.1.198, dated around July 1, 2026.
Here is the useful mental model. Left to its own defaults, Claude will produce a chart, but it makes the generic choices anyone makes without design training: a bar chart where a line would read better, a rainbow palette that falls apart for colorblind readers, ten series crammed onto one axis. /dataviz is a reference sheet that loads the good defaults before Claude writes any code. It does not add a new tool to the model so much as add taste to it.
If the word "skill" is new to you, our primer on what a Claude skill is covers the format from scratch. The short version: a skill is a folder with a SKILL.md file that teaches Claude a task your way. /dataviz is Anthropic teaching Claude how to draw a chart.
How do you use /dataviz?
Two ways, the same two every skill offers. The direct way: type /dataviz in a session, the same as any other slash command. That pulls the full guidance into context on purpose, which is what you want when you are about to build a dashboard and you want the design rules held from the first line of code.
The automatic way: just ask. Community write-ups of the skill (for example Daniel San's changelog thread on X) report that its description triggers on words like "chart," "dashboard," "visualize," and "plot," along with library names like D3 and Recharts. I could not independently verify the exact trigger list from Anthropic's source, so treat that as reported. But the behavior is standard: Claude reads every skill's description at startup and loads the full body when your request matches. That is how skills load, and it is why an installed-but-idle skill costs almost nothing.
A session looks like this. You say "build me a dashboard for these weekly signups." Claude notices the visualization task, loads /dataviz, and its choices change: an encoding that reads accurately, a sane visual hierarchy, and, if the validator does what the changelog says, a palette checked for contrast and colorblind safety before you get the code. You did not ask for any of that. The skill did.
What does /dataviz actually contain?
This is the part worth slowing down for, because /dataviz is a SKILL.md like any other, and reading one that a strong team wrote is the fastest way to learn to write your own.
One honest caveat. The exact text of the bundled skill ships inside the CLI package, and I did not have a confirmed copy of its full body to quote. What follows is the structure every good skill shares, illustrated by what /dataviz is reported to do. Take the specifics as directional.
A skill like this earns its keep in three layers:
- A sharp description. The frontmatter description is the trigger, so it names the task and the moment: charts, dashboards, "visualize," plus the specific libraries a user might mention. A vague description misfires. A specific one fires exactly when it should. This single field does more work than anything else in the file.
- Opinionated guidance, not options. A weak skill lists every chart type and shrugs. A strong one takes positions: choose encodings by perceptual accuracy, never rely on color alone because a meaningful share of readers are colorblind, match the rendering technology to the number of elements on screen. Reported summaries of
/datavizdescribe exactly these kinds of rules. Opinions are the product. If your skill does not decide anything, it is just documentation. - A runnable helper. The color-palette validator is the tell. A skill is a directory, so it can carry a small script that does one thing the same way every time, here checking a palette for contrast and accessibility. That is the line between a prompt and a skill: a prompt can only advise, a skill can ship a tool that runs.
Read those three layers back and you have a checklist for anything you build: a description that triggers precisely, guidance that takes a side, and an optional script for the mechanical part.
How do built-in skills differ from ones you install?
Mechanically, barely at all. That is the point.
Built-in skills like /dataviz, /code-review, /debug, /loop, and /claude-api are prompt-based skills that ship inside the CLI and are available in every session, unless you turn them off with the disableBundledSkills setting. They are the same shape as a skill you drop in .claude/skills/: Anthropic wrote the instructions, and Claude orchestrates the work with its normal tools.
The differences are practical, not structural. Built-in skills live in the package, so they are not yours to edit in place and they update with the CLI. A skill you install lives in your directories, travels with your repo, and changes when you change it. But both are SKILL.md files, both auto-trigger or run on command, both load only when relevant. When Anthropic wanted to give Claude chart taste, they did not build a special subsystem. They wrote a skill. That is the strongest possible endorsement of skills as the extension mechanism for agents, which is the whole bet Knack is built on.
What other skills pair with /dataviz for data work?
/dataviz handles the picture. It does not handle the numbers behind it, and that gap is where you compose: one skill that reasons about data, a second that renders it. For the analysis half, a skill like analytics turns a raw table into the summary worth charting, and ab-test-analysis does the statistical read on an experiment before you visualize the lift. Run one of those, then let /dataviz shape the output into something a stakeholder can read at a glance. If your deliverable is a document rather than a live UI, the built-in document skills for Excel, Word, PowerPoint, and PDF handle the export.
Browse the marketplace for a data skill that fits your pipeline, install it next to the built-ins, and you have a small analysis-to-visual assembly line, each skill single-purpose.
FAQ
Is /dataviz available on all plans?
It ships with Claude Code v2.1.198 and later, so availability tracks having a recent CLI rather than a specific tier. The changelog does not attach /dataviz to a particular plan, so if you are current, you have it.
Can I edit the /dataviz skill?
Not in place. It is a bundled skill that lives inside the CLI package, so it is read-only and updates with the CLI. If you want a variant, copy the pattern into your own .claude/skills/ skill and adjust the guidance there. You can also disable the bundled set with the disableBundledSkills setting.
Does /dataviz work in Cowork?
I could not verify this from Anthropic's docs, so I will not claim it outright. Cowork runs the same agentic engine and reads the same skills, so a bundled skill should be present there too. But bundled skills are documented in the Claude Code context specifically, so confirm it in your own Cowork session before relying on it.
Can I build my own chart-style skill?
Yes, and /dataviz is the template. Write a SKILL.md with a description that names the task and the moment, fill the body with opinionated rules rather than a menu of options, and add a small validator script for the mechanical checks. That is the same three-layer structure the built-in uses.
Where do built-in skills live on disk?
Inside the installed Claude Code package, not in your ~/.claude/skills/ or project .claude/skills/ folders, which is why they are read-only. The exact path depends on how you installed the CLI, so I would not hardcode it. Your own skills stay in the two .claude/skills/ directories.
What chart libraries does /dataviz favor?
Community summaries mention D3 and Recharts among its triggers, which suggests it steers toward the mainstream JavaScript charting libraries. I could not confirm a definitive favored list from Anthropic's source, so treat that as reported, not official. The safest read: the skill is about design principles first and stays library-flexible.