Skills

Open Claude Design: Connect coding agents to Claude Design's visual design workspace

Best suited for doing UI design in Claude Design's visual workspace grounded in a real codebase with two-way sync — provided you're on macOS/Linux/WSL2 with a paid Claude account; remote writes require explicit authorization by default.

Official source · GitHub - maxritter/open-claude-design

Install

curl -fsSL https://github.com/maxritter/open-claude-design/releases/latest/download/install.sh | sh

Who it's for

  • Developers who already have a paid Claude account (Pro, Max, Team, or Enterprise) and use one of 20+ coding agents such as Claude Code, Codex, OpenCode, Cursor, or GitHub Copilot, and want to connect UI design work to a real codebase.
  • Teams that need a two-way workflow: viewing designs in a visual design workspace and syncing design changes back to code (or the reverse).
  • Prerequisites: the OS must be macOS, Linux, or WSL2; Claude Code does not need to be installed. If you don't have a paid Claude account, or your environment isn't one of the above, the README doesn't describe any other way to use it, so availability can't be confirmed.
  • The official evidence doesn't mention restrictions on specific text languages, frameworks, or project structures; defer to the official documentation before use.

First try in 3 minutes

The recommended main path is the official one-line install script:

curl -fsSL https://github.com/maxritter/open-claude-design/releases/latest/download/install.sh | sh
  1. Prerequisite: confirm you're on macOS, Linux, or WSL2 and hold a paid Claude account.
  2. Installation: paste the command above into a terminal and run it. Note this is a pipe-to-shell approach — the install script runs directly on your system; if you're uncomfortable with that, download and review the script first. The installer automatically detects coding agents already installed on your machine and completes the integration. Only use the --all-agents option if you genuinely need every available integration.
  3. Trigger a task: after installation, simply mention Claude Design in a request to your coding agent. The first invocation automatically loads the Claude Design access skill and connects to your design workspace. For your first task, we recommend a low-risk viewing or analysis operation — for example, asking the agent to link the current project's designs with the codebase — rather than directly requesting code changes.
  4. Success signal: the agent reports that the access skill is loaded and it has automatically connected to the design workspace; you can see codebase-linked design content in the Claude Design visual workspace.

Note: review-type diagnostic operations are only proven to be used for diagnostics; whether they read/write files, execute commands, or access the network is not stated in the official evidence, so don't assume they're zero-risk. For sync operations that modify code or files, confirm side effects before acting.

Use cases

Case 1: Connect an existing codebase to the design workspace. Scenario: you want to do UI design based on a real project, not a blank canvas. How: after installation, send a request mentioning "Claude Design" in your coding agent — for example, asking it to load the current project's interface designs into the Claude Design workspace. The real workflow: the agent automatically loads the access skill and connects to the design workspace, directly linking the codebase with the visual workspace. Expected result: codebase-grounded design content appears in the workspace.

Case 2: Sync design changes back to code. Scenario: you've adjusted the interface in the design workspace and want the changes to land in real project files. How: mention Claude Design in an agent request and ask it to sync the design changes to code. The real workflow is two-way design/code sync; remote access is read-only by default, writes require your explicit authorization, and they go through a guarded helper with plan tokens, etag checks, backups, and verification. Expected result: changes are synced after confirmation, with backups available for traceability. If you attempt it without authorization, all you'll see is read-only viewing — no writes will occur.

Usage boundaries

  • File and command access: the CLI install runs a script on your machine; design/code sync involves reading and writing the codebase. Remote access is read-only by default, writes require explicit authorization, and protections include plan tokens, etags, backups, and verification.
  • Network access: connecting to the Claude Design design workspace is an external service call; beyond that, the evidence doesn't describe other network behavior.
  • Confirmation point: any operation that writes design changes into code should have its impact scope confirmed before execution.
  • Environment and account: macOS/Linux/WSL2 + a paid Claude account; Claude Code is not required. Other environments are not documented officially.
  • Explicit official restriction: don't use --all-agents casually — only when you need all integrations.
  • Unconfirmed information: the specific file/command/network behavior of diagnostic operations, the range of supported text languages, and integration details on less common agents are all not stated in the official evidence; consult the official README before use.