Canvas Design Harness for DeepSeek Harness

English | 简体中文

🎨 Interactive Figma-style canvas design plugin & dual-spec Skill for DeepSeek Harness. Create, inspect, and iterate on multi-frame UI designs directly inside DeepSeek Harness Web GUI, while remaining 100% standalone for Codex, Claude Code, and terminal workflows.

📸 Visual Overview

🖼️ 1. Design Gallery (Grid Overview) — Browse all workspace design documents

🎨 2. Multi-Frame Canvas (Grok Mobile App) — Infinite canvas with responsive device frames

📊 3. Living Specs (AI Apps Comparison) — Multi-page design specs version-controlled in Git

🌟 Key Highlights

1. Integrated "Design Gallery" in DeepSeek Harness UI

• Seamlessly mounts into the conversation.view tab ring (alongside Chat and Trajectory).
• Instant visual gallery of all design documents located in docs/designs/.
• Embedded full-height interactive viewer with zoom, pan, and real-time hot-reloading.

2. HTML as Universal Living Specifications

• Human-Friendly: Clean, readable HTML files (docs/designs/*.html). Double-click to open in any standard web browser without heavy proprietary design software.
• AI-Friendly: Standard semantic DOM structure. LLMs can inspect, generate, and perform atomic mutations on frames and components with 100% precision.
• Git-Native Specs: Design files live right in your repository as code. Commit, branch, diff, and review living design specifications alongside your application code.

3. Figma-Like Interactive Canvas Experience

• Multi-page document architecture with infinite pan/zoom canvas.
• Built-in device frame presets (Mobile, Desktop, Modals, Flow diagrams).
• Zero-dependency built-in HTTP-MCP daemon (127.0.0.1:9321) handles live synchronization and SVG exports.

4. Dual-Spec Package

• DeepSeek Harness Plugin: Declares dsh.bundle.patch & dsh.client for complete host-and-client GUI integration.
• Codex / Claude Code Skill: Self-contained skill engine at .agents/skills/canvas-design-harness/, ready to run standalone.

🚀 Quick Start

1. Install into DeepSeek Harness

Option A: Direct Git Install (Recommended)

Install directly from GitHub into your DSH Web profile:

dsh plugin --profile web add git+https://github.com/askman-dev/dsh-canvas-design-harness.git

Option B: Local Clone (For Development)

git clone https://github.com/askman-dev/dsh-canvas-design-harness.git
dsh plugin --profile web add ./dsh-canvas-design-harness

Restart Web Profile

After installation, restart the DSH Web profile in your terminal:

dsh web
# or run the restart script:
python3 /path/to/dsh-canvas-design-harness/scripts/restart-web.py

Refresh your browser (http://127.0.0.1:3080), and you will see the Design Gallery (设计大厅) tab in the top navigation ring.

2. Conversational Design Generation

1. Describe your UI in chat:

   "Design a user login and registration interface for me with mobile screens"

2. Automatic Generation: The Agent invokes canvas_harness_* tools to generate and render design files under docs/designs/.
3. Inspect in Design Gallery: Open the Design Gallery tab to view your interactive canvas.

🏗️ Repository Layout

dsh-canvas-design-harness/
├── .agents/skills/canvas-design-harness/   # Canonical skill engine (synced from upstream)
│   ├── SKILL.md       # Skill definition & guidelines
│   ├── reference/     # Canonical styles & scripts (canvas-frames.css / canvas-frames.js)
│   ├── server/        # Zero-dependency HTTP-MCP daemon (port 9321)
│   └── specs/         # Protocol specifications
├── client/            # DSH Web GUI client module
│   ├── design-view.js # React frontend source (DSH locale & Design System)
│   ├── bundle.js      # Built classic-script bundle (__ModuleLoader__)
│   └── logic.js       # Pure logic & testing helpers
├── host/              # Host tools bridge
│   ├── tools.js       # Model tool bridge definitions (canvas_harness_*)
│   └── tools-entry.js # Cordis tool registration entry
├── scripts/           # Build, verification & maintenance scripts
│   ├── build-client.mjs  # Bundle packaging script
│   ├── restart-web.py    # Sandbox-safe profile restarter
│   ├── verify-web.mjs    # Live GUI acceptance test script
│   └── sync-from-upstream.sh # Upstream skill synchronization
├── docs/assets/       # Visual preview screenshots
├── plugin.js          # DSH plugin entry (skills, daemon lifecycle, /canvas/* routes)
├── package.json       # Manifest declaring dsh.bundle.patch & dsh.client
├── cordis.patch.yml   # Cordis profile patch configuration
└── test/smoke.mjs     # 42-check offline integration test suite (Parts A–F)

🛠️ Model Tools Bridge

The plugin exposes the following structured tools to the AI model (addressed by root + name):

💻 Development & Acceptance

Run the three-step acceptance suite after making any changes:

# 1. Rebuild the client bundle
npm run build:client

# 2. Run the 42-check offline test suite (Skills, Discovery, Daemon, E2E Routes, Boot Race)
node test/smoke.mjs

# 3. Verify against the live running Web GUI
node scripts/verify-web.mjs

Standalone Skill (Codex / Claude Code)

The skill can also run independently without DeepSeek Harness:

cd .agents/skills/canvas-design-harness/server
node src/index.js <your-designs-folder>      # Defaults to http://127.0.0.1:9321

📄 License

MIT License