Web Analyze — Development skill for Claude Code
Analyze live web page content with custom prompts using WebFetch.
How to install Web Analyze
Installs to ~/.claude/commands/33-strategies-33-strategies-claude-code-starter-kit-web-analyze.md
mkdir -p ~/.claude/commands && curl -fsSL https://raw.githubusercontent.com/33-Strategies/33-strategies-claude-code-starter-kit/HEAD/.claude/commands/web-analyze.md -o ~/.claude/commands/33-strategies-33-strategies-claude-code-starter-kit-web-analyze.md Restart Claude Code, or start a new session, for it to be picked up.
What Web Analyze does
description: Analyze live web page content with custom prompts using WebFetch allowed-tools: WebFetch argument-hint: " [optional: specific prompt for what to extract/analyze]" category: workflow
Web Page Analysis with WebFetch
**User Request:** $ARGUMENTS
Instructions
Parse the user's arguments to extract:
- URL - The web page to analyze (required)
- Prompt - What specific information to extract or analyze (optional)
If no specific prompt is provided, use a
Alternatives in Development
- Reddit Fetch — Fetches Reddit content via Gemini CLI when WebFetch is blocked or returns 403 errors 6.5k ★
- Analyze Cluster Postmortem — Postmortem analysis of cluster failures → OPTIMIZE PROMPTS to work for EVERY use case 1.7k ★
- Web Developer Portfolio — Page builder: Elementor 648 ★
Full documentation available on GitHub
View Source RepositoryRelated Skills
Web Research MCP
Read-only Web search + page extraction MCP server (Streamable HTTP). Multi-client: Cursor / Claude Code / Chat
Helpful Content Review Skill
A Claude skill that reviews and scores any draft or live page against Google's helpful, reliable, people-first
Serialised Browsing
use a headless Chromium to fetch a page when curl or WebFetch cannot, in-built burst access prevention. Also a
Claude Yolo
Run parallel Claude Code agents in tmux with automatic permission approval. Auto-approves Bash, WebFetch, and
Display Tab
Create a display tab — a custom HTML page rendered in a deepsteve tab
Wp Page
Generate page templates — blog, generic, legal, 404, search, embed, or custom page types
Related Agents
Domain Extractor
Layer-0 subagent for the rev-3 agentic-ontology methodology. Reads the target project's live documentation (We
10 Content Strategist
AI prompts, landing page copy, email templates, UX microcopy, brand voice. Invoke for any written content or A
Demo Validator Web
Browser-driving validator for the live-demo and workbench builds — uses the Playwright MCP server (configured