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

skill

A skill file that tells your agent when a figure would help and which graph to use. The same two files work in Cursor, Claude Code, Codex, OpenCode and any other agent that supports Agent Skills.

For the bigger picture of how agents write and read these figures, see for agents. The skill doesn't include the graphs themselves, so install the components first if your project doesn't have them yet.

install

A project copy is committed with the repo, so everyone on the team gets it. A personal copy lives only on your machine. Your agent loads the skill when its description matches the task, which is usually when a figure would explain something faster than prose. If your agent reads skills from a different folder, put the same two files there.

project

personal

from the repo

prompt

try it

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

Refactor · GraphFlow, then GraphTimeline

Incident · GraphTimeline, then GraphUptime

Pull request · GraphDiff, then GraphSlope

Pick one · GraphCompare, then GraphRank

README · Fenced ASCII, not JSX

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

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

when to use

Before writing a long list of bullets, the skill checks whether a figure would be quicker to read. If the whole point fits in one sentence, it skips the figure.

The writing isUse
A path or a refactorGraphFlow, then GraphTimeline
An incidentGraphTimeline, then GraphUptime
Pick A vs BGraphCompare, then GraphRank
What a PR changedGraphDiff, then GraphSlope
Overlapping workGraphGantt, then GraphStat
A migration in flightGraphMeter, then GraphKpi
One headline numberGraphKpi, then GraphStat
An RFC or a launch listGraphSheet, then GraphCheck
Nested filesGraphTree
A decision or an ADRDecision, then GraphScore
A roadmap or a sprintGraphBoard, then GraphCheck
A code sample to explainAnnotate, then Terminal
An agent sessionChat, then GraphDiff
Setup or an APIEnv, then Endpoint

For complete write-ups, see examples. To render plain .md in a Comark app, see Comark, and to turn data into Markdown, see Knap.

rules

  • At most two graphs in a section. Prose between them.
  • MDX: markdown children inside the tag, from the docs .mdx tab. Notion / README / GitHub / Linear: the fence from the docs .md tab. Comark: ::graph-* . Knap: graph_* filter. Do not invent ASCII.
  • In MDX, write Markdown inside the tag. Bold is now, italic is next, label: value is a row, and '— note' is a side note. The rules are on /docs/grammar.
  • Titles: short uppercase, drawn as [ TITLE ].
  • Labels: lowercase, plain (auth middleware, not AuthMiddleware Layer).
  • Copy the markdown children from docs or recipes. Paste the official fence when the host cannot render MDX. Props still work when the data is already JavaScript. Do not invent APIs, extra hues, or chart libraries.
  • palette="duo" / "multi" only when a second or third series needs it.
  • Motion is already in the components. Do not add loops or pulses.

These are written for the agent, so they read like a checklist. The short version: never draw a chart as SVG, never make up a text figure, don't restyle the frame, and keep each section to two graphs at most.

the file

This is exactly what the agent loads. /skill.md and /skill/recipes.md always match the repo. If the skill isn't installed, /llms.txt has the same guide along with every format.

skill.md