Skip to content

idonafraid-create/GCW

Repository files navigation

GCW

English · 简体中文

Evidence-driven website teardown, faithful cloning, and creative reconstruction.

GCW — evidence-driven website reconstruction

CI License: MIT Release

GCW is an agent skill and toolkit for reconstructing authorized public websites from runtime evidence. It turns routes, interaction states, responsive behavior, assets, and GPU rendering into an auditable specification before implementation begins.

The result can be a technical teardown, a runnable replay, or an editable faithful clone with a clear path to post-review creative work.

A screenshot is one frame. A live website is a million.

GCW stands for Gao Copy Website. Yes, the name is that literal.

Evidence orchestration: the GCW difference

A screenshot captures one frame. A faithful reconstruction must also explain what happens across routes, breakpoints, hover and focus states, loading transitions, external data, and WebGL or shader pipelines.

GCW keeps three things separate:

  • Evidence — what was actually observed.
  • Faithful implementation — what reproduces the approved baseline.
  • Creative work — what may change only after that baseline passes review.

This separation makes visual similarity testable and prevents a production bundle replay from being mistaken for maintainable source.

Quick start

1. Install

Clone GCW into a stable location, then link it into your agent's skill directory:

git clone https://github.com/idonafraid-create/GCW.git D:\path\to\GCW
New-Item -ItemType Junction `
  -Path D:\path\to\.agent\skills\gcw `
  -Target D:\path\to\GCW

On macOS or Linux, use a symbolic link instead:

ln -s /path/to/GCW /path/to/.agent/skills/gcw

Install and check the toolkit:

npm ci
npm run install:browser
python -m pip install -r requirements.txt
npm run check

Requirements: Node.js 20+, Python 3.10+, Chromium installed through Playwright, and design-dna installed at .agent/skills/design-dna. Design DNA is required for every teardown depth, including minimal. For Canvas/WebGL/WebGPU/shader targets, also install web-shader-extractor at .agent/skills/web-shader-extractor and run npm run check:gpu.

2. Ask for an outcome

Examples:

Use GCW to analyze this authorized site and produce a technical teardown only.

Use GCW to build an editable faithful clone of this authorized site.

Use GCW to build an editable faithful clone, review it with me, then continue with approved creative changes.

Before build work starts, GCW asks you to choose the final delivery contract. It does not choose for you:

Choice Final deliverable Editability target
A Research or runnable replay only Runnable replay
B Editable faithful clone Maintainable source
C Editable faithful clone, then Creative after review Maintainable source

Choosing B does not lock you out of Creative. You can explicitly upgrade to C at the review gate or resume Creative later from an accepted B delivery.

What you get

GCW evidence, reconstruction, and verification

Depending on scope, GCW leaves behind:

  • a finalized SITE_SPEC.md tied to captured evidence;
  • route, interaction, responsive, asset, network, and GPU inventories;
  • source-versus-local visual regression results;
  • an implementation decision trail with provenance and known gaps;
  • for B/C, a maintainable source entrypoint, REPLACE_GUIDE.md, and editability evidence.

ARTIFACT_REPLAY may be used as an oracle, but it cannot pass a B/C delivery gate as the final candidate.

How it works

TEARDOWN_PHASE -> FAITHFUL_CLONE -> REVIEW_GATE -> CREATIVE_REBUILD
  1. Observe routes, states, assets, rendering systems, and reuse constraints.
  2. Specify the baseline and label conclusions as source, partial, or guess.
  3. Reconstruct through source adaptation, a clean rebuild, or authorized production recovery.
  4. Verify builds, routes, responsive states, runtime independence, and visual diffs before review.

When production recovery is required, MAINTAINABLE_REBUILD is the formal strategy for turning recovered evidence into maintainable source.

For visual-system analysis, GCW integrates design-dna. GPU targets additionally require web-shader-extractor. Their native evidence remains authoritative; GCW coordinates it rather than duplicating it.

Documentation

Need Start here
Full agent workflow SKILL.md
Clone modes and A/B/C contracts references/clone-modes.md
Phase and delivery gates references/gates.md
Site specification references/site-spec.md
Recovery strategies references/recovery-tiers.md
Runtime independence references/runtime-independence.md
Tool and script reference references/tooling.md
QA scenarios references/qa-scenarios.md

Run the complete repository gate with:

npm run verify

Safety and reuse

Use GCW only for sites you own, license, or have explicit authorization to reconstruct. Private or authenticated content is out of scope unless separately authorized. Code, fonts, images, models, data, and branding must each be reviewed for reuse rights.

GCW records evidence and uncertainty; it does not turn deployed artifacts into presumed original source.

Acknowledgments

GCW builds on ideas and tooling from design-dna, web-shader-extractor, and Playwright.

License

MIT

About

Evidence-driven website teardown, faithful cloning, and creative reconstruction, powered by Design DNA and WebGL forensics.

Topics

Resources

License

Stars

5 stars

Watchers

0 watching

Forks

Packages

 
 
 

Contributors