SKILL.md · diff
git:20261014.0344ebd to git:20261015.13d6774
22 added, 28 removed. Audit C to C.
# Use MuseDraw
- MuseDraw is a professional drawing assistant delivered as an Agent Skill. Its first
- public workflow produces component diagrams in **手绘线稿** and **终端色域**.
+ MuseDraw is a professional drawing skill for AI agents: research figures,
+ office diagrams, teaching visuals and system architecture, in 20 visual styles.
To use this repository for a drawing task, read
- [the MuseDraw skill](skills/musedraw/SKILL.md), then load only the references and
- helpers relevant to the request. Keep the complete `skills/musedraw/` directory
- available; the entrypoint alone is not the whole workflow.
-
- Short natural-language requests are sufficient to start; the agent handles
- composition, palette and layout. The two public styles share installed drawing
- code and accept new subjects, components and relationships; the showcased topics
- do not limit the workflow.
-
- For these diagrams, read the [semantic graph workflow](skills/musedraw/references/graph-workflow.md).
- The agent writes semantic content and invokes the installed renderer; it does
- not ask the user for coordinates or draw a case-specific replacement SVG.
- Follow-up content edits operate on the existing document. Deliver PNG, native
- editable SVG and `document.json`, with the checks actually completed.
+ [the MuseDraw skill](skills/musedraw/SKILL.md) and keep the whole
+ `skills/musedraw/` directory available (engine, themes, fonts manifest,
+ references). The fastest way to learn the scene format is to copy the closest
+ case in [gallery/](gallery/README.md); each case has `scene.json` and a
+ one-line `prompt.md`.
- Replaying a saved semantic document with the same drawing version and fonts
- is distinct from generating a new diagram from a short request. Preserve that
- distinction; neither the examples nor their verification guarantee identical
- output from every model. See [implementation and limits](docs/semantic-graph.zh-CN.md).
+ The user's words decide the content. Gallery examples are references, not facts
+ about the user's project; numbers in them are illustrative. Creating a diagram
+ is not permission to change the user's application, publish their data or
+ install unrelated software. First-time setup (`npm install` in the skill folder)
+ installs open-source fonts and Playwright and attempts to download Chromium.
+ It requires the corresponding network access and installation permission.
- The user's brief determines the content. Repository examples are generic
- references, not facts about their system. Creating or updating a diagram is not
- permission to change the user's application, publish their data, or install
- unrelated software.
+ Use `scripts/musedraw.mjs render scene.json` in the installed skill for new work;
+ `scripts/musedraw.mjs themes` lists the 20 styles. Deliver PNG and editable SVG,
+ retain the scene for follow-up edits, open the PNG and address `checks.json`
+ warnings. With `--svg-only`, state that PNG and browser checks are incomplete.
- For installation and host requirements, see [中文使用指南](docs/usage.zh-CN.md)
- or [English usage](docs/usage.md). The host needs file access and tool execution;
- PNG uses existing Node.js 22+, Playwright, Chromium-compatible browser and fonts.
- MuseDraw itself requires no additional model API key.
+ Replaying an existing scene in the same engine, theme, font and rendering
+ environment is distinct from a model interpreting a new request. Do not promise
+ pixel-identical output or universal model success from a short prompt. Old small
+ component-graph tools remain compatibility routes, not the main gallery engine.
+ See [usage](docs/usage.md) or [中文使用指南](docs/usage.zh-CN.md).