A slash-command style website cloning workflow for Claude Code / Claude Desktop command setups.
This package keeps the cloning pipeline in one shared context while strengthening the inputs:
- full raw HTML
- stripped blueprint HTML for navigation
- external CSS extraction
- branding metadata
- screenshot metadata
- optional screenshot QA loop
commands/clone.mdcommands/clone-setup.shcommands/extract-css.shcommands/smart-strip.sh
Use the installer:
bash install.shOr copy the files into your Claude commands directory manually:
mkdir -p ~/.claude/commands
cp commands/clone.md ~/.claude/commands/clone.md
cp commands/clone-setup.sh ~/.claude/commands/clone-setup.sh
cp commands/extract-css.sh ~/.claude/commands/extract-css.sh
cp commands/smart-strip.sh ~/.claude/commands/smart-strip.sh
chmod +x ~/.claude/commands/clone-setup.sh
chmod +x ~/.claude/commands/extract-css.sh
chmod +x ~/.claude/commands/smart-strip.shnpxfirecrawl-clicreate-next-app- Playwright support in your Claude environment for visual QA
If Firecrawl is not authenticated:
npx firecrawl-cli login --browser/clone https://example.com
/clone https://example.com my-project --qa
/clone https://example.com my-project --skip-scaffold
/clone --from-existing ~/proj-example
The repository includes a reusable QA prompt at prompts/qa-review-prompt.md for screenshot-based fix iterations after the initial build.
GitHub Actions runs a minimal sanity check on every push and pull request:
- shell syntax validation for the helper scripts
- installer dry run against a temporary
HOME
The pipeline deliberately avoids splitting the page into isolated section-generation agents. The current approach keeps a single translator agent in full-page context and uses helper scripts to improve source fidelity instead of reducing shared context.