name: website-cloner
description: Build an improved website clone from a URL via 6-phase gated workflow (Vite/React/shadcn/Tailwind + GitHub Pages). Use for end-to-end site rebuilds. Don't use for single-phase work.
license: MIT
effort: high
metadata:
    version: 1.2.1
    author: "Luong NGUYEN <luongnv89@gmail.com>"

Website Cloner

6-phase orchestrator that clones any website and produces an improved version — better performance, UI/UX, SEO, and security — built with Vite + React + shadcn/ui + Tailwind CSS, deployable to GitHub Pages.

When to Use

Trigger when the user asks to:

• Clone, rebuild, or recreate a website ("clone this site", "make a better version of <url>")
• Improve a website's performance, UI/UX, or SEO by analyzing and rebuilding it
• Start a full website improvement workflow from a URL

Do not use for single-phase work (use the specific sibling skill directly).

Prerequisites

• A target URL (publicly reachable preferred).
• Write access to create a local project dir (defaults under ~/workspace/clones or $CLONE_DIR).
• The sibling phase skills installed (website-analyzer, website-clone-report, etc.).
• Optional: GitHub token if you want Pages deploy automation in Phase 5.
• User approval at gates (explicit confirmation before Phases 3, 4, 5 advance).

If a prerequisite is missing, stop and report it — do not guess paths or credentials.

Workflow

Phase 1 — Analyze        → website-analyzer
Phase 2 — Report (gate)  → website-clone-report
Phase 3 — Propose (gate) → website-improvement-prd  (outputs prd.md)
Phase 4 — Plan  (gate)   → website-implementation-plan  (outputs tasks.md)
Phase 5 — Build          → website-builder
Phase 6 — Final Report   → website-clone-final-report

Approval gates after Phase 2, 3, and 4: the orchestrator must not advance without explicit user approval.

Artifacts (each written once, then referenced by name in the phases below): analysis.json — Phase 1's structured findings; report.md — Phase 2's plain-language summary; prd.md — Phase 3's improvement proposal; tasks.md — Phase 4's phased implementation plan, including the approved GitHub Actions artifact-deployment task; builder-metadata.json — Phase 5's build metadata, workflow-produced Pages URL, and structured post-deployment performance/SEO/security snapshot consumed by Phase 6; after-analysis.json — the comparable Phase 5 re-audit source. Phase 5 also produces base-aware Vite configuration and .github/workflows/deploy-pages.yml, which builds and deploys dist/ rather than publishing the repository root.

Layout

This umbrella and its phase skills live together in a single suite folder:

skills/website-cloner/                    ← this umbrella
├── SKILL.md                              ← orchestrator (you are here)
├── website-analyzer/                     ← Phase 1
├── website-clone-report/                 ← Phase 2
├── website-improvement-prd/              ← Phase 3
├── website-implementation-plan/          ← Phase 4
├── website-builder/                      ← Phase 5
└── website-clone-final-report/           ← Phase 6

Why nested: the phases are tightly coupled to this umbrella's data flow (analysis JSON → report → PRD → tasks → built site → final report). Keeping them in one folder makes the suite easy to browse, audit, and ship together. Each phase skill stays independently installable — the installers (install.sh, remote-install.sh) discover both top-level and nested skills.

When invoking phase skills below, refer to them by name (/website-analyzer, /website-clone-report, …); the runtime resolves names regardless of filesystem path.

See the individual phase skill docs for their full references/ and scripts/. This orchestrator stays short to fit the agent's context budget.

Repo Sync Before Edits (mandatory)

Before modifying files in a repository:

branch="$(git rev-parse --abbrev-ref HEAD)"
git fetch origin
git pull --rebase origin "$branch"

If dirty, stash first:

git stash push -u -m "website-cloner: pre-sync"
branch="$(git rev-parse --abbrev-ref HEAD)"
git fetch origin && git pull --rebase origin "$branch"
git stash pop

If origin is missing or rebase conflicts occur, stop and ask.

Setup

1. Resolve working directory — the directory where the cloned website will be built. If $CLONE_DIR is set, use it. Otherwise ask the user once and save to ~/.config/website-cloner-dir.txt. Default: ~/workspace/clones.
2. Create project folder under resolved root: YYYY_MM_DD_<slug_from_url>/
3. Set $PROJECT_DIR to the created folder path.
4. If no URL provided in $ARGUMENTS, ask the user for one.

Phase 1: Understand the Website

Invoke website-analyzer with the URL:

/website-analyzer <url> --output "$PROJECT_DIR/analysis.json"

The analyzer produces a structured analysis covering: UI/UX, category, style, performance (lcp_estimate_seconds, unitless cls_estimate, ttfb_estimate_seconds, page weight in KB, and request count), surface-level security, and SEO (overall score + per-dimension breakdown).

Check: Analysis file exists and covers all 6 dimensions. If any dimension is missing, note it but continue — partial results are acceptable for Phase 2.

Step Completion Report:

◆ Analyze (step 1 of 6 — <site name>)
······································································
  UI/UX profile:        √ pass | × partial — <gap>
  Category detected:    √ pass (<category>)
  Style analysis:       √ pass | × partial — <gap>
  Performance metrics:  √ pass (LCP=<val> CLS=<val>)
  Security surface:     √ pass | × partial — <gap>
  SEO score:            √ pass (<score>/100)
  ____________________________
  Result:               PASS | PARTIAL

Phase 2: End-User Report (GATE)

Invoke website-clone-report with the analysis output:

/website-clone-report "$PROJECT_DIR/analysis.json" --output "$PROJECT_DIR/report.md"

This skill produces a plain-language report for non-technical readers and prompts for approval before persisting. The orchestrator waits for user approval here.

Check: report.md exists and was approved by the user.

Step Completion Report:

◆ Report (step 2 of 6 — <site name>)
······································································
  Report written:       √ pass (report.md)
  User approved:        √ pass | × pending
  ____________________________
  Result:               PASS | BLOCKED

If not approved, do not advance. Ask the user to review the report and approve or request changes.

Phase 3: Improvement Proposal (GATE)

Invoke website-improvement-prd with the approved report and analysis:

/website-improvement-prd "$PROJECT_DIR/report.md" "$PROJECT_DIR/analysis.json" --output "$PROJECT_DIR/prd.md"

This skill produces a full improvement proposal with what/why/value for each change and writes prd.md after user approval.

Check: prd.md exists and was approved.

Step Completion Report:

◆ Proposal (step 3 of 6 — <site name>)
······································································
  prd.md written:       √ pass
  User approved:        √ pass | × pending
  Changes catalogued:   √ pass (<N> proposed changes)
  ____________________________
  Result:               PASS | BLOCKED

Phase 4: Implementation Plan (GATE)

Invoke website-implementation-plan with prd.md:

/website-implementation-plan "$PROJECT_DIR/prd.md" --output "$PROJECT_DIR/tasks.md"

This skill produces a phased implementation plan with landing page first, asset collection vs. creation, individual tasks, and a deterministic GitHub Actions Pages artifact deployment from base-aware Vite dist/ output — written to tasks.md after user approval.

Check: tasks.md exists and was approved.

Step Completion Report:

◆ Plan (step 4 of 6 — <site name>)
······································································
  tasks.md written:     √ pass
  User approved:        √ pass | × pending
  Phases defined:       √ pass (≥ 2 phases)
  Landing page first:   √ pass
  Pages artifact task:  √ pass (base-aware dist/)
  ____________________________
  Result:               PASS | BLOCKED

Phase 5: Build

Invoke website-builder with tasks.md and prd.md:

/website-builder "$PROJECT_DIR/tasks.md" "$PROJECT_DIR/prd.md" --output "$PROJECT_DIR/"

This skill executes the plan, builds the site (Vite + React + shadcn/ui + Tailwind), and deploys the verified dist/ artifact through .github/workflows/deploy-pages.yml. The workflow sets the Vite base to / for a user/organization Pages repository or /<repo>/ for project Pages; repository-root and branch-folder publishing are forbidden. It then re-runs website-analyzer against the responsive workflow-produced Pages URL and emits the structured after snapshot in metadata for Phase 6. If deployment or the re-audit is incomplete, Phase 5 must return PARTIAL and preserve nulls/errors rather than inventing metrics.

Step Completion Report:

◆ Build (step 5 of 6 — <site name>)
······································································
  Landing page built:   √ pass
  Assets collected:     √ pass (<N> assets)
  Assets created:       √ pass (<N> assets)
  Pages dist artifact:  √ pass | × unavailable
  GitHub Pages URL:     √ pass (<url>) | × unavailable
  After metrics:        √ complete | × partial ([performance/SEO/security gaps])
  ____________________________
  Result:               PASS | PARTIAL | FAIL

Phase 6: Final Comparison Report

Invoke website-clone-final-report with the analysis baseline and builder metadata:

/website-clone-final-report "$PROJECT_DIR/analysis.json" "$PROJECT_DIR/builder-metadata.json" --output "$PROJECT_DIR/final-report.md"

This skill validates the baseline and structured post-deployment snapshot, then produces a before/after comparison covering performance, SEO, security, UI/UX changes, deviations, and the GitHub Pages URL. It must propagate PARTIAL when any required performance, SEO, or security comparison is unavailable.

Step Completion Report:

◆ Final Report (step 6 of 6 — <site name>)
······································································
  final-report.md:      √ pass
  Performance delta:    √ pass (before→after) | × partial ([missing])
  SEO delta:            √ pass (before→after) | × partial ([missing])
  Security comparison:  √ pass (before→after) | × partial ([missing])
  Deviations listed:    √ pass | × none
  ____________________________
  Result:               PASS | PARTIAL | FAIL

Expected Output

◆ Website Cloner — <site name>
┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄
  Phase 1  Analyze      √ pass
  Phase 2  Report       √ approved
  Phase 3  Proposal     √ approved (prd.md)
  Phase 4  Plan         √ approved (tasks.md)
  Phase 5  Build        √ pass | × partial
  Phase 6  Final Report √ pass | × partial | × fail
  Overall Result:       PASS | PARTIAL | FAIL

  GitHub Pages: https://<user>.github.io/<repo>/
  Project:      <project_dir>

The overall result is the worst Phase 1–6 result. Never report overall PASS when the Phase 5 after snapshot or any Phase 6 required comparison is partial, unavailable, or failed.

Edge Cases

• Unreachable URL: Stop Phase 1, report error, do not continue. Ask user for a different URL.
• JS-heavy SPA with no crawlable content: Note the limitation in Phase 1, proceed with best-effort analysis. The build phase may need user-supplied assets to compensate.
• User denies approval at any gate: Stop the pipeline. Do not auto-proceed. The user can re-run the skill to resume.
• Missing sibling skill: If a sibling skill is not available, skip that phase and note it in the report. Continue where possible, but the overall result cannot exceed PARTIAL.
• Partial Phase 1 results: If the analyzer returns partial data (e.g., no SEO score), proceed to Phase 2 with a note; Phase 6 and the overall workflow remain PARTIAL when a required baseline comparison is unavailable.
• Partial Phase 5 re-audit: Continue to Phase 6 so it can document the gaps, but both Phase 6 and the overall workflow must return PARTIAL unless report creation itself fails.