For AI agents
An AI assistant can learn how to use document-svg from this site without reading any code. Here is how to give it the tool, and what it should keep in mind.
Give your assistant the tool
Claude Code
Add the marketplace and install the plugin. You get commands to convert, write back, tidy up SVG, make PNG previews and check the setup.
/plugin marketplace add ryusui-hiro/document-svg
/plugin install document-svg@document-svgCodex
Run the installer from a local checkout, then start a new task. The same skill becomes available.
./scripts/install-codex-plugin.shGitHub Copilot
Add the setup action to your Copilot setup steps and the coding agent can run the command while it works.
What an agent should know
- For ordinary files, use the simple convert step. The lower-level building blocks are for people drawing SVG from their own shapes.
- The file extension decides how a file is read. Check the format list before assuming a type is supported.
- Every page, slide or sheet becomes its own SVG, and one report covers the whole job.
- Finishing without an error does not mean perfect. Warnings mean a person should take a look.
- Use a new, empty output folder for every conversion.
- Use only the published interface. The internal readers can change between releases.
- Don't raise the safety limits to force a difficult file through. They protect the machine from oversized or hostile files.
- Turning an SVG back into Office or CAD keeps how it looks, not the original text flow, cells or formulas.
A summary written for language models
llms.txt is a short summary of the project for AI assistants: what it does, how to call it, and which links to read next.
The format list as data
The format list on this site is also published as a data file, in English, Japanese and Chinese, so an assistant doesn't have to read the web page.