Skill · llms.txt · MDX
For agents
When a write-up needs a figure, the agent already has one. A skill file picks the graph. JSX goes in MDX. The official fence goes in a README, a PR, or Linear.
Write and read
Writing — the agent emits at most two graphs next to the claim. React gets JSX. Plain Markdown gets the official twin from /llms.txt.
Reading — the figure is characters in the file. Opening the MDX shows labels and values, not a screenshot. The agent can edit the frame the same way it wrote it.
- MDX / React✓–
- README–✓
- GitHub / Linear–✓
- PR comment–✓
- Edit the labels✓✓
What you drop in
33 graphs. Most have an official ASCII twin for files that cannot run React. The skill picks at most two.
- Skill
- When to use a figure, which one, and whether to write JSX or paste the official fence. Same two files in Cursor, Claude Code, Codex, or OpenCode.
- Recipes
- Worked write-ups with real props. Copy the JSX, swap the labels. At most two graphs, prose between them.
- Twins
- Fenced ASCII that survives GitHub, Linear, and a PR comment. Swap labels, keep the frame. Do not invent a fence.
- llms.txt
- Chooser plus twins in one file, if the skill is not installed.
- /skill.md
- the skill
- /skill/recipes.md
- jsx recipes
- /llms.txt
- chooser + twins
- /developers
- api + openapi
- at most two graphs
- prose between them
- official twin, not homemade
- no SVG
Install the skill
Project copy travels with the repo. Personal copy is this machine only. If the host is React, install the components first. The graphs themselves still need to be in the project.
Project
Personal
From the repo
Prompt
Chooser table and the full file: Skill.
Try it
Paste one of these after install. Each should pick two graphs and put prose between them.
Refactor · GraphFlow, then GraphTimeline
README · Fenced twins, not JSX