How to Customize the AI Website Cloner Pipeline for Different Fidelity Levels

You can customize the AI website cloning pipeline for different fidelity levels by passing inline instructions at invocation time, editing the default scope in .github/skills/clone-website/SKILL.md, or creating reusable fidelity profile files under docs/research/fidelity-profiles/.

The JCodesMore/ai-website-cloner-template repository provides a multi-phase cloning pipeline that defaults to pixel-perfect fidelity. By adjusting the clone-website skill definition and its execution parameters, you can tailor the workflow to generate anything from full interactive replicas to rapid visual mockups.

Understanding the Default Fidelity Configuration

The cloning behavior is governed by .github/skills/clone-website/SKILL.md. This file contains a Scope Defaults section that sets the default fidelity to pixel-perfect, ensuring exact matches in colors, spacing, typography, and animations.

The Pre-Flight section (lines 27-34) checks for optional user instructions, allowing runtime overrides without file modifications. The standard pipeline executes in distinct phases: reconnaissance, global token extraction, interaction sweep, component specification, parallel builders, assembly, and visual QA.

Method 1: Pass Custom Instructions at Invocation

The fastest way to adjust fidelity is to append instructions when invoking the skill. The Pre-Flight logic explicitly honors user-provided instructions over the defaults defined in the skill file.

/clone-website https://example.com
Fidelity level: high-level mockup
Skip interaction sweep, generate only layout components

The agent reads this additional context and automatically skips the exhaustive interaction sweep and component specification phases.

Method 2: Edit the Default Fidelity in SKILL.md

For persistent changes across all cloning operations, modify the source skill definition directly.

Open .github/skills/clone-website/SKILL.md and locate the Scope Defaults block. Change the fidelity descriptor and comment out or remove phases you do not need:

## Scope Defaults

...
- **Fidelity level:** High-level mockup — layout and core styling only

After saving your changes, propagate them to all platform-specific skill copies by running:

node scripts/sync-skills.mjs

This script regenerates the instruction sets for all agent copies based on your modified SKILL.md.

Method 3: Create Reusable Fidelity Profiles

If you frequently switch between fidelity levels, create standardized YAML profiles under docs/research/fidelity-profiles/. This approach keeps your configurations portable and version-controlled.

Create a file named quick-preview.yaml:

fidelity: quick-preview
phases:
  - reconnaissance
  - foundation

Then reference this profile during invocation:

/clone-website https://example.com --profile docs/research/fidelity-profiles/quick-preview.yaml

The agent parser reads the profile and conditionally executes only the listed phases.

Pipeline Adjustments by Fidelity Level

Different fidelity levels trigger specific phase skip logic in the cloning pipeline:

• Pixel-perfect (default): All phases execute including full Interaction sweep and Component Specification & Dispatch. Builders receive detailed specs for exact replication.

• High-level mockup: Skips the exhaustive Interaction sweep and Component Specification. Only global fonts, colors, and basic layout screenshots are captured. Builders receive simplified layout-only instructions.

• Quick-preview: Runs only Phase 1 – Reconnaissance (full-page screenshots) and a minimal Foundation step that creates a single globals.css with primary colors. No component files are generated.

• Low-detail: Retains the Foundation step but omits all asset downloads; images are replaced with placeholders. This is optimal when you need only a visual scaffold for rapid prototyping.

Key Files for Customization

• .github/skills/clone-website/SKILL.md: Defines the default fidelity, scope defaults, and the full multi-phase pipeline. This is the primary file for understanding what gets extracted in each phase.

• scripts/sync-skills.mjs: Regenerates platform-specific skill copies after any modification to the source skill file. Must be executed after editing SKILL.md.

• AGENTS.md: Serves as the single source of truth for all agent instructions. Modify this file when adding new fidelity-related commands or behaviors.

• docs/research/fidelity-profiles/: Optional directory for storing reusable fidelity configuration files for quick switching between different project requirements.

Summary

• Inline instructions provide immediate, per-invocation fidelity adjustments without modifying repository files.
• Editing SKILL.md creates persistent default changes for all future cloning operations.
• Fidelity profiles enable standardized, reusable configurations for different project needs.
• Always run node scripts/sync-skills.mjs after modifying .github/skills/clone-website/SKILL.md to update platform-specific copies.
• Lower fidelity levels skip resource-intensive phases like interaction sweeps and asset downloads, significantly reducing execution time while sacrificing detail.

Frequently Asked Questions

Can I switch between fidelity levels without editing the SKILL.md file?

Yes. Pass a custom instruction block immediately after the URL when invoking the skill. The Pre-Flight section (lines 27-34) in .github/skills/clone-website/SKILL.md automatically detects and honors these overrides, allowing you to specify "high-level mockup" or "quick-preview" on a per-job basis without permanent configuration changes.

What happens if I remove phases from the SKILL.md file?

Removing or commenting out phases in the Scope Defaults section permanently disables those steps for all subsequent cloning operations. For example, deleting the "Interaction sweep" section prevents the agent from capturing hover states and click behaviors, reducing output fidelity but improving execution speed.

How do I propagate changes to all platform-specific agents?

After modifying .github/skills/clone-website/SKILL.md, execute node scripts/sync-skills.mjs from the repository root. This script regenerates the instruction sets for all platform-specific agent copies, ensuring consistency across your deployment environment according to the repository's synchronization logic.

Is there a way to preview what phases will run before executing the clone?

While there is no built-in dry-run flag, you can create a quick-preview.yaml fidelity profile under docs/research/fidelity-profiles/ that lists only the phases you want to validate. Invoke the skill with --profile pointing to this file to test the pipeline logic before running a full pixel-perfect clone.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}