whiteboard-diagrams
Create accurate, simple, hand-drawn whiteboard diagrams from a user's process, system, architecture, learning workflow, comparison, or causal explanation. Use when the user asks for a scribbly whiteboard visual, educational workflow graphic, hand-drawn explainer, or diagram like the bundled reference; ask one focused clarification question when the subject or relationships are missing, then generate and visually verify the final image.
- License 许可
- MIT
- 最近更新
- 2026-11-14
- Stars
- ★ 0
- Forks
- 1
01 前言 Front Matter
- name
- whiteboard-diagrams
- description
- Create accurate, simple, hand-drawn whiteboard diagrams from a user's process, system, architecture, learning workflow, comparison, or causal explanation. Use when the user asks for a scribbly whiteboard visual, educational workflow graphic, hand-drawn explainer, or diagram like the bundled reference; ask one focused clarification question when the subject or relationships are missing, then generate and visually verify the final image.
02 资源文件 10
- 参考文档AGENTS.md1.9 KB
- 参考文档agents/openai.yaml226 B
- 参考文档examples/text-to-image-model.json1.3 KB
- 许可证LICENSE1.0 KB
- 参考文档README.md2.2 KB
- 参考文档references/EXAMPLES.md1.5 KB
- 参考文档references/INTAKE.md1.5 KB
- 参考文档SCALE.md9.8 KB
- 脚本scripts/diagram_prompt.py6.6 KB
- 脚本tests/test_diagram_prompt.py2.4 KB
03 SKILL.md 正文
Whiteboard Diagrams
Turn a described system into one causal picture that a new reader can follow without narration.
Workflow
-
Read
SCALE.mdcompletely before composing or generating anything. -
Extract the subject, audience, components, relationships, required labels, and output format from the request.
-
Apply the intake gate below. Do not invent a missing workflow.
-
Convert the request into a compact diagram spec. Use
references/INTAKE.mdfor the schema andreferences/EXAMPLES.mdwhen the topology is unclear. -
Validate and compile the spec:
python scripts/diagram_prompt.py validate path/to/spec.json python scripts/diagram_prompt.py build path/to/spec.json --output /tmp/diagram-prompt.txt -
Generate one raster image with the available image-generation tool using the compiled prompt. For a brand-new diagram, do not pass unrelated reference images. If the user supplies a style reference, label it as a style reference, never as an edit target.
-
Inspect the actual output at full size. Check every required label, arrow, merge, lane, margin, and distinction against the source spec.
-
Make one targeted correction when text, causality, or clipping is wrong, then inspect again. Do not call an inaccurate diagram finished.
-
Return the image and, for project-bound work, its saved path.
Intake gate
Proceed without asking when the request provides both:
- what the diagram must explain; and
- the main steps, components, or relationships.
If either is missing, ask exactly one concise question and stop generation until answered:
What should the diagram explain, and what are the main steps or components and how do they connect?
Do not ask about style, palette, or dimensions unless the user explicitly needs a non-default treatment. Use the defaults in SCALE.md for those choices.
Non-negotiable rules
- Preserve causal truth. An attractive wrong arrow is a failed output.
- Use the user's exact technical names and verbatim copy.
- Show state changes, joins, feedback loops, and frozen versus trainable parts explicitly.
- Keep the whiteboard human: loose marker strokes, imperfect arrows, terse labels, and small explanatory doodles.
- Keep the information disciplined: no fake UI, decorative cards, gradients, glows, shadows, icon tiles, background grids, or ornamental clutter.
- Assign color by meaning, not decoration. Default to black structure, blue flow, and one red emphasis for the learned or critical component.
- Keep all live content clear of the canvas edge. Nothing may be clipped.
- Never hide missing information behind generic labels such as "process," "data," or "AI magic."
- Never claim generation succeeded without inspecting the produced image.
Output contract
The final answer must include:
- the generated diagram;
- a one-sentence description of what it explains; and
- the saved path when the image belongs to a project or repository.
Do not expose the full internal prompt unless the user requests it.
04 同仓库更多 Skills
zarr-python
Chunked N-D arrays for cloud storage (Zarr-Python 3). Compressed arrays, parallel I/O, S3/GCS via fsspec, NumPy/Dask/Xarray compatible, for large-scale scientific computing pipelines.
whiteboard-diagrams
Create accurate, simple, hand-drawn whiteboard diagrams from a user's process, system, architecture, learning workflow, comparison, or causal explanation. Use when the user asks for a scribbly whiteboard visual, educational workflow graphic, hand-drawn explainer, or diagram like the bundled reference; ask one focused clarification question when the subject or relationships are missing, then generate and visually verify the final image.
webapp-testing
Toolkit for interacting with and testing local web applications using Playwright. Supports verifying frontend functionality, debugging UI behavior, capturing browser screenshots, and viewing browser logs.
web-artifacts-builder
Suite of tools for creating elaborate, multi-component claude.ai HTML artifacts using modern frontend web technologies (React, Tailwind CSS, shadcn/ui). Use for complex artifacts requiring state management, routing, or shadcn/ui components - not for simple single-file HTML/JSX artifacts.
verification-loop
A comprehensive verification system for Claude Code sessions.
vercel-deploy
Deploy applications and websites to Vercel. Use when the user requests deployment actions like "deploy my app", "deploy and give me the link", "push this live", or "create a preview deployment".