What problem does it solve? AI-generated website clone analyses are often full of hallucinated code that breaks when run. This Skill enforces a real-source-first methodology with executable probes so you can faithfully reproduce any website, from static pages to WebGL-heavy interactive demos, without trusting fabricated code. ## Core Features & Use Cases - 6-Step Decision Tree: Search GitHub for real source first, then probe the site with browser recon scripts, pick a path (wget mirror, template rebuild, static-site asset mirroring, or WebGL reverse-engineering), build the project, verify in a real browser, and replace content with your own. - Executable Probe Scripts: Includes recon-site, mirror-site, route-crawl, interaction-probe, network-capture, asset-harvest, sourcemap-hunt, visual-diff, compare-recon, audit-clone, and dna-scaffold scripts for evidence-based cloning. - Evidence Discipline for WebGL: Applies SOURCE/PARTIAL/GUESS evidence grading, a no-compensation rule, and a baseline-first replay gate when reverse-engineering Canvas/Three.js effects. - Use Case: Clone a WebGL demo site by mirroring every deployed asset (including runtime-fetched .wasm/.buf files), then generate a CLONE_REPORT.md with pixel-diff scores comparing original and clone. ## Quick Start Ask the AI to clone a website by saying "复刻这个网站: https://example.com" or "Clone this site for me" and it will load the decision tree and run the recon scripts.
