Skip to content
mdxcn is now part of Shadcn Labs. details

write a figure your agent can edit

When a write-up needs a figure, the skill tells your agent which graph to use and how to write it for the file at hand: JSX in MDX, a ::graph-* block in a Comark app, a graph_* filter in Knap, or a plain-text figure in a README, pull request or Linear note.

pnpm dlx shadcn@latest add @mdxcn/all
[ loop ]

writing and reading the same figure

When writing, the agent adds at most two graphs, each next to the point it supports. When it comes back later, the figure is still plain text in the file, so it can read the labels and values directly instead of squinting at a screenshot.

GitHub, Linear and READMEs don't run Comark, so they get the plain-text figure.

[ host ]
JSXASCIIComarkKnap
  • MDX / React✓–✓–
  • App .md file––✓✓
  • README / GitHub–✓–✓
  • Edit the labels✓✓✓✓
[ loop ]
write-up
chooser
JSX / YAML / filter
open the file
read the frame
edit labels

what goes in the agent's folder

There are 47 graphs, and most have a plain-text version for places that can't run React. Here's what the agent gets to work with.

skill
Tells the agent when a figure helps, which graph fits, and which format the file needs: JSX, a ::graph-* block, a graph_* filter or the plain-text figure. The same two files work in Cursor, Claude Code, Codex and OpenCode.
recipes
Complete write-ups with real data. The agent copies one, swaps in its own labels, and keeps to two graphs with prose in between.
comark
::graph-* blocks with YAML props, for apps that render plain Markdown with Comark instead of MDX.
knap
graph_* filters that turn data into the plain-text figure inside a Knap template, or into a ::graph-* block for Comark.
text figures
Code blocks that look the same in GitHub, Linear and pull request comments. The agent edits the labels and leaves the frame alone.
llms.txt
The graph guide and every format in a single file, for agents that don't have the skill installed.
[ kit ]
  • skills/mdxcn
  • SKILL.mdchooser
  • recipes.mdjsx
  • fetch
  • /llms.txtascii + yaml
  • /docs/exampleswrite-ups
Tree with 6 nodes
[ rules ]
  • at most two graphs
  • prose between them
  • official fence, not homemade
  • no svg
4 of 4 done

install the skill

A project copy is committed with the repo; a personal copy stays on your machine. If your site runs React, install the components first, since the graphs have to be in the project to render.

The full skill file and the graph guide are on the skill page.

project

personal

from the repo

prompt

try a prompt

After installing, paste one of these prompts. A good result uses two graphs, with prose in between.

Refactor · GraphFlow, then GraphTimeline

README · Fenced ASCII, not JSX

Comark · ::graph-timeline, then ::graph-uptime

Knap · {{ events | graph_timeline }}, then {{ uptime | graph_uptime }}