Agent Skills: tldraw-creator Skill

tldraw-creator Skill

Create and manage tldraw whiteboards programmatically, deploy self-hosted instances, and build collaborative whiteboarding workflows.

Overview

tldraw-creator enables agents to:

• Create whiteboards programmatically using the @tldraw/store API (shapes, text, connectors, frames)
• Deploy self-hosted instances via Docker or Node.js with PostgreSQL persistence
• Import/export boards as .tldraw JSON documents
• Manage board galleries and enable collaborative editing
• Visualize diagrams directly in whiteboards with custom shapes and layouts

Prerequisites

• Node.js 18+ (for programmatic board creation)
• Docker (optional, for self-hosted server deployment)
• PostgreSQL 12+ (optional, for persistent document storage)
• npm (for installing @tldraw packages)

Setup

Install tldraw Packages

Note: As of tldraw 3.x all tldraw functionality is bundled in the single tldraw package. The individual @tldraw/store, @tldraw/tlschema, @tldraw/ui, and @tldraw/editor sub-packages are still available but the tldraw meta-package is the recommended entry point.

# Recommended: single package (tldraw 3.x)
npm install tldraw react react-dom

# Optional: individual sub-packages for advanced use
npm install @tldraw/store @tldraw/tlschema
npm install pg yjs y-websocket  # For server + persistence
npm install express cors         # For API server (optional)

Vite / ESM Setup

tldraw ships dual CJS/ESM code. When using Vite, add lodash.isequalwith to optimizeDeps.include to avoid a missing-default-export error at runtime:

// vite.config.js
import { defineConfig } from 'vite'
import react from '@vitejs/plugin-react'

export default defineConfig({
  plugins: [react()],
  optimizeDeps: {
    include: ['lodash.isequalwith', 'lodash.isequal'],
  },
})

Environment Variables

Create .env file in your skill workspace root or pass as process variables:

# Programmatic Board Creation
TLDRAW_EXPORT_FORMAT=json          # json | svg | png (default: json)

# Self-Hosted Server
TLDRAW_SERVER_PORT=3000
TLDRAW_DATABASE_URL=postgresql://user:password@localhost:5432/tldraw
TLDRAW_WEBSOCKET_URL=wss://tldraw.example.com
TLDRAW_NODE_ENV=development        # development | production

# Optional: Cloud Storage
TLDRAW_S3_BUCKET=my-tldraw-boards
TLDRAW_S3_REGION=us-east-1

Tool Calling Pattern

Tool: Create Board

Purpose: Programmatically generate a tldraw whiteboard with shapes, text, and connectors.

Input Schema:

{
  "name": "my-board",
  "title": "Project Architecture",
  "shapes": [
    {
      "id": "shape-1",
      "type": "geo",
      "x": 100,
      "y": 100,
      "props": {
        "w": 200,
        "h": 100,
        "geo": "rectangle",
        "color": "blue",
        "richText": { "type": "doc", "content": [{ "type": "paragraph", "content": [{ "type": "text", "text": "Frontend" }] }] }
      }
    }
  ],
  "theme": "dark"
}

tldraw 3.x: shape labels use richText (ProseMirror document), not text. Use toRichText(str) from the tldraw package to convert plain strings.

Output Schema:

{
  "success": true,
  "boardId": "doc_abc123",
  "boardName": "my-board",
  "exportPath": "/path/to/my-board.tldraw",
  "boardJson": { "/* TLStore snapshot */" }
}

Implementation: See scripts/create-board.js

Tool: Deploy Server

Purpose: Set up a self-hosted tldraw instance for visualization and collaboration.

Input Schema:

{
  "deploymentType": "docker",
  "persistenceType": "postgresql",
  "databaseUrl": "postgresql://tldraw:password@localhost:5432/tldraw",
  "serverPort": 3000,
  "enableWebSocket": true
}

Output Schema:

{
  "success": true,
  "serverUrl": "http://localhost:3000",
  "status": "running",
  "containerName": "tldraw-server",
  "logs": "Server started..."
}

Implementation: See scripts/deploy-server.js

Tool: Import Board to Server

Purpose: Upload a board JSON to a running self-hosted instance and get a visualization URL.

Input Schema:

{
  "serverUrl": "http://localhost:3000",
  "boardJson": { "/* TLStore snapshot */" },
  "boardName": "my-board"
}

Output Schema:

{
  "success": true,
  "boardId": "doc_abc123",
  "viewUrl": "http://localhost:3000/room/doc_abc123",
  "editUrl": "http://localhost:3000/edit/doc_abc123"
}

Implementation: See scripts/import-board.js

Tool: List Boards

Purpose: Get a list of all boards on a self-hosted instance (for gallery/management).

Input Schema:

{
  "serverUrl": "http://localhost:3000",
  "limit": 20,
  "sortBy": "created_at"
}

Output Schema:

{
  "success": true,
  "boards": [
    {
      "id": "doc_abc123",
      "name": "Project Architecture",
      "createdAt": "2026-04-26T12:00:00Z",
      "updatedAt": "2026-04-26T14:30:00Z",
      "url": "http://localhost:3000/room/doc_abc123"
    }
  ]
}

Implementation: See scripts/list-boards.js

Usage Examples

Example 1: Create and Deploy a Simple Diagram

User Request: "Create a diagram with 3 interconnected components and deploy it to a whiteboard"

Agent Flow:

1. Agent calls create-board with:
   - 3 geo shapes (rectangles)
   - 2 arrow connectors
   - Labels for each component

2. Agent calls deploy-server (if not already running)

3. Agent calls import-board-to-server with the generated board

4. Agent returns URL: http://localhost:3000/room/doc_abc123

Example 2: Create and Save a Flow Diagram

User Request: "Build a swimlane diagram for a user onboarding flow"

Agent Flow:

1. Agent designs shapes:
   - Frames for each swimlane (user, backend, email service)
   - Text boxes for steps
   - Arrows showing flow direction

2. Agent calls create-board

3. Board is saved as JSON to workspace: /my-swimlane-flow.tldraw

4. Agent can export to PNG/SVG if needed

Example 3: Build and Manage a Board Gallery

User Request: "Show me all my whiteboards and let me pick one to edit"

Agent Flow:

1. Agent calls list-boards on running server

2. Agent generates gallery UI with:
   - Board names and creation dates
   - Thumbnail previews
   - Links to view/edit each board

3. User selects a board → opens in tldraw viewer

Skill Scripts

scripts/create-board.js

Programmatically create a .tldraw file with shapes, bindings, and text.

Usage:

node scripts/create-board.js \
  --name "my-diagram" \
  --title "System Architecture" \
  --config config.json

Output: my-diagram.tldraw (JSON serialized TLStore snapshot)

See references/api-reference.md for detailed shape API.

scripts/deploy-server.js

Deploy a self-hosted tldraw instance using Docker or Node.js.

Usage:

node scripts/deploy-server.js \
  --type docker \
  --port 3000 \
  --database postgresql://localhost/tldraw

Output: Docker container running tldraw server, accessible at http://localhost:3000

See references/deployment-guide.md for detailed setup.

scripts/import-board.js

Import a board JSON into a running tldraw server.

Usage:

node scripts/import-board.js \
  --server http://localhost:3000 \
  --board my-diagram.tldraw \
  --name "My Diagram"

Output: Public URL to view/edit the board on the server.

scripts/list-boards.js

List all boards on a running server (for gallery management).

Usage:

node scripts/list-boards.js \
  --server http://localhost:3000 \
  --format json

Output: JSON array of board metadata.

Shape Types Reference

tldraw supports a rich set of shape types. See references/shape-reference.md for complete details.

| Type | Example | Typical Use | |------|---------|------------| | geo | Rectangle, Ellipse, Diamond, Triangle | Flowchart nodes, diagrams | | text | Text box | Labels, annotations, notes | | arrow | Connector with arrowhead | Flow direction, relationships | | frame | Container/artboard | Grouping shapes, swimlanes | | image | Embedded image | Diagrams with images, photos | | note | Sticky note shape | Quick notes, reminders | | bookmark | Web link preview | References to external content |

Deployment Guide

Quick Start: Self-Hosted with Docker

1. Install Docker: https://docs.docker.com/get-docker/

2. Run setup script:

   node scripts/deploy-server.js --type docker
   

3. Access: Open http://localhost:3000 in browser

4. Create boards and import:

   node scripts/create-board.js --name my-diagram
   node scripts/import-board.js --server http://localhost:3000 --board my-diagram.tldraw
   

Production Deployment

See references/deployment-guide.md for production considerations:

• Nginx reverse proxy with SSL/TLS
• WebSocket configuration
• PostgreSQL persistence
• Environment variable setup
• Health checks and monitoring

API Reference

Document Serialization Format

tldraw documents are JSON files with the .tldraw extension. Structure:

{
  "version": 15,
  "document": {
    "id": "doc_abc123",
    "pages": {
      "page_abc123": {
        "id": "page_abc123",
        "name": "Page 1",
        "shapes": { "/* shape objects */" },
        "bindings": { "/* binding objects */" }
      }
    },
    "assets": { "/* images, videos */" },
    "pageStates": { "/* UI state */" }
  }
}

Key Points:

• Documents can have multiple pages
• Shapes are keyed by ID in a dictionary
• Bindings define connections between shapes
• Assets store embedded media (base64 encoded)

See references/api-reference.md for complete schema and type definitions.

Known Breaking Changes (tldraw 3.x)

text → richText (tldraw 3.0+)

All shape props that previously accepted a plain string for label text now require a ProseMirror document object (richText). This affects both geo shapes and text shapes.

| Shape type | Old prop (< 3.0) | New prop (≥ 3.0) | |-----------|-----------------|------------------| | geo | props.text | props.richText | | text | props.text | props.richText | | arrow | props.text | props.richText |

Use the toRichText(str) helper exported by tldraw to convert plain strings:

import { toRichText } from 'tldraw'

// Single line
toRichText('Hello world')
// → { type: 'doc', content: [{ type: 'paragraph', content: [{ type: 'text', text: 'Hello world' }] }] }

// Multi-line (one paragraph per line)
function rt(str) {
  return {
    type: 'doc',
    content: str.split('\n').map(line => ({
      type: 'paragraph',
      content: line.length ? [{ type: 'text', text: line }] : [],
    })),
  }
}

align → textAlign for text shapes

text shape horizontal alignment prop was renamed:

// Old (< 3.0)
{ type: 'text', props: { align: 'middle' } }

// New (≥ 3.0)
{ type: 'text', props: { textAlign: 'middle' } }

Prefer editor.createShapes() over loadSnapshot()

store.loadSnapshot() requires exact schema sequence numbers that change with every release. Prefer the editor API for programmatic board creation — it is version-independent:

// ✅ Reliable: works across tldraw versions
<Tldraw onMount={(editor) => {
  editor.createShapes([
    { id: createShapeId('box'), type: 'geo', x: 100, y: 100,
      props: { w: 200, h: 100, geo: 'rectangle', richText: toRichText('Hello') } }
  ])
  editor.zoomToFit()
}} />

// ⚠️  Fragile: schema versions must match installed tldraw version exactly
store.loadSnapshot({ schema: { schemaVersion: 2, sequences: { ... } }, store: { ... } })

Arrow labels: always bring to front

Text labels placed next to arrows are independent shapes. Create them after arrows, then call editor.bringToFront() so they always render on top regardless of creation order:

editor.createShapes(labelShapes)
editor.bringToFront(labelShapes.map(s => s.id))

Layout hardening rules (required)

Use these rules for every generated board to prevent common rendering regressions:

1. Normalize escaped line breaks before converting to rich text.

• Convert \\n to real newlines (\n) first.
• Then call toRichText(normalizedText).

2. Generate KPI cards from a computed grid, not hand-typed coordinates.

• Use constants: baseX, baseY, cardW, cardH, colGap, rowGap.
• Build each card rectangle from (row, col).

3. Run overlap assertions before creating shapes.

• Assert KPI cards do not overlap each other.
• Assert connector labels do not overlap section rectangles.
• Fail fast (throw) in development if any overlap is detected.

4. Place connector labels in gutters between sections.

• Do not place labels inside dense content blocks.
• Keep horizontal and vertical gutters explicitly reserved in layout constants.

5. Bring labels to front after shape creation.

• Call editor.bringToFront() for connector labels and badges.

6. Reserve safe margins for editor chrome.

• Keep important badges/cards at least 220px away from the right edge to avoid overlap with the tldraw style panel.
• Keep key content at least 60px from top/bottom edges for toolbar and zoom/minimap controls.

.tldraw file format vs. skill input schema

assets/example-architecture.tldraw in this skill uses a simplified input schema designed for human/agent readability — it is not a native tldraw store snapshot. Do not pass .tldraw files from this skill directly to store.loadSnapshot(). Instead, parse them and create shapes via editor.createShapes().

A working reference implementation is in test-server/src/App.jsx.

Troubleshooting

Server Won't Start

• Check Node.js version: node --version (must be 18+)
• Check Docker installation: docker --version
• Check port availability: lsof -i :3000 (macOS/Linux) or netstat -an | findstr 3000 (Windows)

Database Connection Failed

• Verify PostgreSQL is running
• Check connection string in .env
• Test with: psql $DATABASE_URL -c "SELECT 1"

Board Import Failed

• Ensure server is running: curl http://localhost:3000/health
• Verify board JSON format matches schema
• Check server logs: docker logs tldraw-server (if using Docker)

WebSocket Connection Refused

• Check Nginx is configured for WebSocket upgrade
• Verify Upgrade and Connection headers are proxied
• Test with: wscat -c wss://tldraw.example.com

Resources

• Official tldraw Repo: https://github.com/tldraw/tldraw
• Programmatic API Docs: https://tldraw.dev/docs/introduction
• tldraw Store Package: https://www.npmjs.com/package/@tldraw/store
• Yjs/y-websocket: https://docs.yjs.dev/
• PostgreSQL Node.js Client: https://node-postgres.com/

Next Steps

1. Review references/api-reference.md for detailed API documentation (includes tldraw 3.x richText schema)
2. Check references/deployment-guide.md for production setup
3. Examine scripts/ for implementation examples
4. See test-server/src/App.jsx for a working self-hosted board using the editor API
5. See assets/example-architecture.tldraw for the simplified skill input schema (not a native tldraw snapshot)
6. Integrate with agent workflows using the tool-calling patterns above

For questions or contributions, refer to research notes in session workspace: /research/tldraw-agent-skill-research.md