fatmind

Webclaw3 — AI skill for Claude Code

AI community

Browser automation & web scraping for AI agents — drives your own logged-in Chrome, then distills one successful run into a free, deterministic skill you can rerun daily.

How to install Webclaw3

This entry records only its repository, not the path inside it, so there is no exact command to give. Open fatmind/webclaw3 and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Webclaw3 does

Browser automation & web scraping for AI agents — drives your own logged-in Chrome, then distills one successful run into a free, deterministic skill you can rerun daily. Claude Code / Codex / Qoder compatible Agent Skill.

Alternatives in AI

Firecrawl MCP Server — 🔥 Official Firecrawl MCP Server - Adds powerful web scraping and search to Cursor, Claude and any other LLM c 6.1k ★
Cursor Tools — by eastlondoner - Creates a versatile AI command interface supporting multiple providers and models with flexi 4.7k ★
Design Md Chrome — Chrome extension to extract styles from any website and generate DESIGN.md files and design skills for AI base 2.7k ★

README

webclaw3 — Browser Automation & Web Scraping for AI Agents

**English** | [中文](./README.zh-CN.md)

Teach your AI agent browser tasks once — run free forever.

**It runs in *your* Chrome, with *your* logins.** No fresh headless browser to re-authenticate, no passwords handed to anyone, no login wall blocking the page you need. Setup is 3 steps, about 5 minutes.

Every time you ask an AI agent to do something in a browser, you pay in tokens — and the result is flaky: it works today, fails tomorrow with nothing changed. webclaw3 turns one successful exploration into a reusable **skill**, so a task you run every day stops costing you every day.

![intro](assets/intro.gif)

The cost problem

A single "go look at this site" task costs millions of tokens. Here's a real API bill for one browser task:

![cost](assets/cost.png)

Prompt 2.3M × $3/MTok = $6.90 | Completion 20.5K × $15/MTok = $0.31 | Cache 2.1M × $0.30/MTok = $0.63
With cache-hit variance: ~$2–$3+ per task on Sonnet
On Opus ($5/$25) the same task runs $5–$8+; complex multi-step exploration easily exceeds $10

Run it daily and that's **$90+ a month** — for the exact same steps, re-thought from scratch every time. On top of that, you've got 3 everyday headaches:

It burns cash daily. Same site, same steps, millions of tokens every single time — like hiring someone who forgets everything overnight.
It's unreliable. AI browsing is inherently flaky; a task succeeds now and fails next time just because elements loaded in a different order.
It needs babysitting. You pay AND watch it run, ready to re-run at any moment.

How it works

The fix: **explore once, distill it into a skill, rerun it for free.** Everything happens on your own machine, inside the agent you already use (Claude Code, workbuddy, qoderwork).

**① Explore — figure out what you actually want.** Just point in a rough direction; the agent drives your real browser with this skill. Requirements are discovered by tr