About this skill

Problem

Many web tasks are not simple HTML fetches: pages may require login state, clicks, scrolling, lazy loading, iframe / Shadow DOM, or evidence inside images and video. Plain WebFetch or curl handles public static content, but fails on anti-bot walls, session parameters, and state that only appears after interaction.

How it works

The skill treats browsing as a loop: define success, choose a starting path, validate each step, and stop when the target is reached.

Light tasks: use WebSearch to locate sources, WebFetch to extract text, or curl for raw structured fields.
Heavy tasks: connect to local Chrome / Edge via a CDP Proxy, then run /new, /navigate, /click, /eval, /scroll, and /screenshot in background tabs while preserving logged-in state.
Site memory: read references/site-patterns/{domain}.md before acting, and record verified URL patterns, platform quirks, or anti-bot traps after success.

Boundaries

It is useful for logged-in workflows, interactive research, media inspection, internal systems, and sites where static fetching is blocked. It is overkill for a single public query. CDP use assumes Node.js 22+, browser debugging enabled, and local Proxy calls that bypass proxies; the skill should leave the user's existing tabs untouched and close only tabs it created.

Screenshots

Use Cases

Open an authenticated admin console in the user's browser and extract key fields from rendered page data.
Inspect article or documentation pages, query image and video URLs from the DOM, then download media directly.
Verify official announcements, policy text, or tooling docs instead of relying on secondary coverage.
Split independent research targets across sub-agents that browse parallel background tabs and return summaries.

Best For

Engineers who need to access logged-in consoles or internal systems and extract rendered page data.
Product managers who need to verify official sites, policy clauses, or product documentation.
AI agent developers who need DOM structures, media URLs, or video frame inspection.
Engineers who need controlled web automation workflows in a local browser environment.

Related Skills

Grill Me Plus

@user_c31fa39f/grill-me-plus

Run a grilling session to interact with or test AI agents.

AI Agent 2026.08.30 ↓ 3.5K

Prompt Alchemist

@user_83d7fe7a/prompt-alchemist

A systematic prompt optimization skill that refines prompts using a four-step distillation framework (diagnose, structure, think, compress) and methodologies from four prompting masters.

AI Agent 2026.08.30 ↓ 3.8K

Tencent Cloud Question Marking Agent

@tencent-adm/tencentcloud-ocr-questionmarkagent

Tencent Cloud OCR agent for K-12 exam grading with async full-sheet or single-question marking, handwritten answer recognition, correctness, knowledge points, and coordinates.

AI Agent 2026.08.30 ↓ 2K

HalfAI Workbench

@user_60ba4767/halfai-gzt

Quickly converts a user's input, list, or screenshot into a multi-page workbench, supporting template selection, custom builds, and responsive layouts.