Gdd Live

Community

hegemonart

gdd-live

Live in-browser design mode. The user picks a DOM element on a running dev server (via the Claude Preview MCP), the agent generates N design variants in one batch, they hot-swap in place through HMR or preview_eval using a data-gdd-variant marker, the user accepts or discards, and the whole pick-generate-accept loop persists to .design/live-sessions so it survives a crash or resume. Use when the user wants to iterate on the look of a live component against a real running server, asks to try variants on a page, or runs the live command with a url; falls back to a screenshot-only degraded mode on harnesses without MCP support. Activates for requests involving in-browser design iteration, picking an element on a dev server, or generating variants with hot-swap.

Overview

Publisher	hegemonart
Repository	`get-design-done`
Skill name	`gdd-live`
Stars	3
Forks	1
Bundled files	Instructions only
License	MIT
Links	Skill source Repository

Markdown instructions
A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.
Works with any LLM
AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.
Self-contained
Everything the model needs lives in the instructions — no extra files to sync.
Open source
Published by hegemonart on GitHub. Read the source before you install it.

Installation

Install the Gdd Live AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

Open the app and go to Plugins → Skills.
Choose "Install from GitHub".
Paste the skill folder URL below and confirm.
https://github.com/hegemonart/get-design-done/tree/main/scripts/skill-templates/live
Enable the skill in any chat where you want it available.

Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills

git clone --depth 1 https://github.com/hegemonart/get-design-done.git /tmp/get-design-done
mkdir -p .claude/skills
cp -r /tmp/get-design-done/scripts/skill-templates/live .claude/skills/gdd-live

Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Gdd Live in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Gdd Live on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Gdd Live is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

Loaded only when it is needed
The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.
Switch models mid-chat
Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

gdd-live - Live In-Browser Design Mode

Pick a DOM element on a running dev server, generate competing design variants, hot-swap them in place, and accept the winner as a real source edit. Every step persists to .design/live-sessions/<id>.json so the session survives a crash or a later resume.

The browser-side runtime, the harness-mode gate, the session store, the events feed, the post-check, the scope guard, and the bandit feed are all separate modules under scripts/lib/live/. This skill describes the loop and names the module that owns each step; it does not import them.

For the full surface (the Preview MCP tools, the six live_* events, the session file, the bandit feed, degraded mode, the scope guard), see ../../reference/live-mode-integration.md. For the SKILL.md structural contract, see ../../reference/skill-authoring-contract.md.

Arguments

[url] - the page to drive. Optional. When omitted, detect the dev server and use its root.
--variants N - how many variants to generate per pick. Default 3.
--resume <session-id> - reattach to an in-progress session in .design/live-sessions/.

BOOT

Probe the Preview MCP per ../../connections/preview.md: ToolSearch({ query: "Claude_Preview" }), then mcp__Claude_Preview__preview_list. Empty ToolSearch means the MCP is not loaded.
Resolve the harness live mode. The capability signal is capability_matrix.mcp_support in scripts/lib/manifest/harnesses.json, projected by scripts/lib/live/harness-mode.cjs (liveModeFor(harnessId)). A puppeteer result means full live mode; a degraded result means screenshot-only.
If mcp_support is false for this harness, or Preview is unavailable, enter DEGRADED mode and say so plainly: variants are generated and captured as static screenshots, with no in-page hot-swap. Skip the INJECT and PICK steps; generate against the file the user names instead.
Detect the dev server. Look for Vite, Next, Bun, or a static server (check package.json scripts plus a preview_list entry). Record the server descriptor on the session.
Open or create the session via scripts/lib/live/session-store.cjs (.design/live-sessions/<id>.json). On --resume, load the named session (see RESUME).

INJECT

Inject the browser runtime once. Read RUNTIME_JS from scripts/lib/live/runtime.cjs and evaluate it in the page with mcp__Claude_Preview__preview_eval. The runtime is an idempotent IIFE bound to window.__gddLive, so a re-inject after navigation rebinds the same singleton rather than stacking listeners. It installs the pick handler and the variant-swap helpers, and stamps the live variant on the element via the data-gdd-variant attribute.

PICK

Arm the picker (window.__gddLive.pick()), then guide the user to click the target element. Use preview_click and preview_inspect to confirm the element and read its computed styles and bounding box.
Read the pick report back. Its fields are documented in pickReportShape (selector, tagName, classList, boundingRect, computedStyle subset, current variant). The selector strategy prefers id, then a data-testid, then a tag plus class plus nth-of-type path.
Emit a live_pick event through scripts/lib/live/events.cjs and append a pick entry to the session.

GENERATE (one batch)

Load the relevant Phase 45 canonical reference index FIRST, so variants are grounded in real guidance: the domain index that matches the picked element (for example ../../reference/spatial.md for layout, ../../reference/interaction.md for components and a11y, ../../reference/color.md for color, ../../reference/typography.md for type, ../../reference/motion.md for animation).
Generate all N variants in ONE batch (default 3), each a distinct, hypothesis-tagged design direction for the picked element. Do not generate them one at a time.
For each variant: write the change atomically to the implicated source file, then make it live. With HMR running, the file write is enough; otherwise apply the variant in place with window.__gddLive.swapVariant({ n, style, html }), which sets data-gdd-variant="n" and applies the variant's style or markup.

POST-CHECK

Run the post-check on each variant via scripts/lib/live/postcheck.cjs, which invokes gdd-detect. Show the findings inline next to each variant. A variant that trips a finding is flagged, NOT auto-rejected: the user still decides. Append a live_postcheck event per variant.

ACCEPT / DISCARD

ACCEPT one variant: apply the chosen variant as the canonical source edit, and revert the others in the page (window.__gddLive.revert() on each non-chosen element). Emit a live_accept event and feed the outcome to the design-variants bandit via scripts/lib/live/bandit-feed.cjs (a dev-time signal). Append an accept entry.
DISCARD: revert every variant in the page back to its captured original and leave the source untouched. Emit a live_discard event and append a discard entry.

Either way, persist the result through scripts/lib/live/session-store.cjs before continuing.

PERSIST

Every step (boot, pick, generate, post-check, accept, discard) is written to the session file through scripts/lib/live/session-store.cjs as it happens. The on-disk event log uses the pick, generate, accept, discard kinds; the telemetry stream uses the six live_* event types. Writes are atomic, so an interrupted step never leaves a half-written session.

RESUME

With --resume <session-id>, load the named session from .design/live-sessions/. Only an in_progress session is resumable. Offer the user two choices: continue from the last recorded event (report what that was, for example "last pick was the primary button"), or start fresh (open a new session and leave the old one intact). Never silently replay completed events.

SCOPE GUARD

Never write outside the source files implicated by the picked element. Run every proposed write through scripts/lib/live/scope-guard.cjs, which maps the picked selector to its owning source files and rejects edits that fall outside them. If a variant would need a change beyond that scope (a shared token, a parent layout, a new dependency), stop and surface it to the user rather than widening the blast radius.

Constraints

Do NOT edit files outside the picked element's implicated sources (enforced by the scope guard).
Do NOT generate variants one at a time; generate the full batch, then swap.
Do NOT auto-reject a variant on a post-check finding; flag it and let the user decide.
In DEGRADED mode, state up front that hot-swap is unavailable and fall back to screenshots.
Persist before every user-facing prompt so a crash never loses accepted work.

LIVE COMPLETE

Frequently asked questions

What does the Gdd Live AI skill do?

Why use Gdd Live on TypingMind?

Because you install it once and use it with any model. Gdd Live is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Gdd Live in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/hegemonart/get-design-done/tree/main/scripts/skill-templates/live. TypingMind reads its SKILL.md and installs it as a skill you can enable per chat.

Which AI models can use Gdd Live?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Gdd Live?

As many as you like. As long as a model supports skills, you can use Gdd Live with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Gdd Live AI skill free?

Yes. It is published on GitHub by hegemonart under the MIT license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Tldraw Offline

NousResearch

OrganizationPopular

Drive and script tldraw offline canvases with an agent.

Brand Guidelines

Applies Anthropic's official brand colors and typography to any sort of artifact that may benefit from having Anthropic's look-and-feel. Use it when brand colors or style guidelines, visual…

Solid

Write, port, review, or test Solid 2 UI in OpenClaw (Control UI, plugin views, Canvas hosts), including state ownership, Lit interop during the migration, native overlays, testing, performance, and…

Canvas

Present hosted widget documents on a connected macOS panel and control panel visibility or navigation.

Overview

Installation

Install in TypingMind

Install in another agent

Use it in TypingMind

Works with any AI model

Skill instructions

gdd-live - Live In-Browser Design Mode

Arguments

BOOT

INJECT

PICK

GENERATE (one batch)

POST-CHECK

ACCEPT / DISCARD

PERSIST

RESUME

SCOPE GUARD

Constraints

LIVE COMPLETE

Frequently asked questions

Related AI Skills

Set up your own AI workspace now