NLR-2007

explaindraw

Create clear visual explanations from natural-language requests as mascot illustrations, editable PowerPoint diagrams, or hybrid illustration-plus-diagram slides. Use when a user asks to draw, illustrate, visualize, make a flowchart, architecture diagram, timeline, roadmap, conceptual analogy, educational visual, or hybrid explanatory slide, especially when they provide this repository URL and ask the AI to use it as a skill.

NLR-2007 1 Updated 1w ago

Resources

12
GitHub

Install

npx skillscat add nlr-2007/illustrations

Install via the SkillsCat registry.

SKILL.md

ExplainDraw

Turn a user's topic into a finished visual artifact using the repository's white-background, hand-drawn visual language. Work from the installed skill directory; do not clone the repository again when these files are already available.

Resolve the repository

  1. Treat the directory containing this SKILL.md as the repository root.
  2. Use the local copy when present. Never delete and reclone a usable local checkout.
  3. If the user supplied a repository URL but no local files exist, clone it once into a writable temporary or workspace directory.
  4. If cloning fails because DNS or network access is unavailable, continue from any existing checkout. If no checkout exists, request a repository archive or restored network access; do not pretend the skill was loaded.
  5. Run commands from the repository root. If node_modules/ is absent, run npm install before using the CLI. On Windows PowerShell, use npm.cmd instead of npm when script execution policy blocks npm.ps1.

Choose one output mode

Respect an explicit user choice. Otherwise choose the smallest mode that communicates the concept clearly; do not force hybrid output.

Build the artifact

Illustration

  1. Convert the request into a concise scene: one visual metaphor, one focal action, only essential objects, and ample negative space.

  2. Preserve the mascot and style locks exactly. Use reference images from references/mascot/ when the active image-generation tool accepts image references.

  3. Prefer the host AI's available image-generation tool to create the actual bitmap. Ask it for a flat pure-white background, black sketch lines, and yellow #FFC21A accents only.

  4. Save the generated image in output/<output-name>/ when the tool returns a writable artifact.

  5. When direct image generation is unavailable, create the deterministic prompt package instead:

    npm run generate -- --input <request.json>

    State clearly that this fallback produced generation-ready prompts, not a rendered image.

Editable diagram

  1. Translate the user's content into diagram JSON matching src/schemas/diagram-schema.ts. Keep labels short and preserve the actual logic; never invent system relationships merely to fill space.

  2. Generate native editable PowerPoint shapes:

    npm run diagram -- --input <diagram.json>
  3. Deliver the .pptx from output/<input-name>/. Do not rasterize editable diagrams unless the user explicitly asks for a preview image.

Hybrid

  1. Create request JSON matching src/schemas/request-schema.ts with mode set to hybrid.

  2. Generate the split visual:

    npm run hybrid -- --input <request.json>
  3. If an image-generation tool is available, render the illustration from the generated prompt and place or return it with the editable PowerPoint. Otherwise deliver the .pptx plus prompt package and identify the illustration placeholder.

Natural-language requests

The user does not need to author JSON. Infer safe defaults and write a small temporary JSON input in the workspace:

  • audience: GENERAL
  • purpose: presentation
  • format: 16:9
  • editable: true
  • outputName: a lowercase, filesystem-safe slug

Ask a question only when a missing choice would materially change the artifact. Otherwise proceed and mention the chosen mode.

Validate before delivery

  1. Run npm test after code or schema changes.

  2. For prompt packages, run:

    npm run validate -- --input output/<output-name>
  3. For diagrams, verify every connection references a real node, no shapes overlap, all content stays inside slide bounds, and text is legible.

  4. For illustrations, visually inspect the rendered result when possible. Reject extra colors, gradients, shadows, scenery, malformed mascot anatomy, missing two yellow chest marks, and accidental text.

  5. Report the actual artifact paths and distinguish rendered images, editable decks, and prompt-only fallbacks.

Non-negotiable visual rules

  • Use a pure white #FFFFFF background, black/near-black linework, and yellow #FFC21A accents only.
  • Keep illustration lines thin, imperfect, and hand-drawn.
  • Draw the mascot as one continuous egg/potato-shaped body, never a snowman-shaped separate head and torso.
  • Keep exactly two small organic yellow chest marks.
  • Keep diagrams editable and do not flatten them into images by default.
  • Avoid decorative complexity that weakens the explanation.