do-wdr-visual-resolver

name	do-wdr-visual-resolver
description	Visual URL resolution skill using CLIP screenshot embeddings to handle scanned PDFs, image-heavy layouts, and JS-heavy SPAs.
license	MIT
compatibility	Python 3.11+, CLIP, Playwright
allowed-tools	Bash(python:*|do-wdr:*) Read
metadata	{"author":"d-oit","version":"0.1.0","source":"https://github.com/d-oit/do-web-doc-resolver"}

Visual Resolver Skill

Extends the Web Doc Resolver cascade with a visual provider based on CLIP screenshot embeddings.

When to use this skill

Activate this skill when text-based extractors (Jina, Firecrawl, etc.) fail to resolve a URL, particularly for:

• Scanned PDFs (raster text)
• Single Page Applications (SPAs) returning skeleton HTML
• Infographic-heavy or chart/table-only documents
• Complex multi-column academic layouts

Prerequisites

Install visual resolution dependencies:

pip install torch torchvision torchaudio
pip install git+https://github.com/openai/CLIP.git
pip install playwright
playwright install chromium

Integration

This skill is designed to be integrated into the scripts/_url_resolve.py cascade.

Usage

from .visual_resolver import VisualResolver

resolver = VisualResolver()
if resolver.is_available():
    result = resolver.resolve("https://example.com/scanned-pdf", "technical architecture diagram")