# Browse

> Browser QA and dogfooding workflow using direct agent-browser commands.

- Skill: `zhongpei/browse` (Agent Skill)
- Install (CLI): `npx skillmds@latest add zhongpei/browse`
- Raw SKILL.md: https://api.skillmd.com/api/skills/zhongpei/browse/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: zhongpei (https://skillmd.com/u/zhongpei)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/zhongpei/browse

---


# Browse: QA Testing & Dogfooding

Use this skill for browser-visible verification, screenshots, and interaction checks.

## Setup

- `agent-browser` must be installed
- `agent-browser install` completed

## Core QA Patterns

### Verify page load
```bash
agent-browser open <url>
agent-browser snapshot --interactive
agent-browser console
agent-browser errors
agent-browser network requests
```

### Test a user flow
```bash
agent-browser open <url>
agent-browser snapshot --interactive
agent-browser fill <selector-or-ref> "value"
agent-browser click <selector-or-ref>
agent-browser snapshot --interactive
```

### Visual evidence
```bash
agent-browser screenshot --annotate
agent-browser screenshot --full
```

### Responsive check
```bash
agent-browser set viewport 375 812
agent-browser screenshot mobile.png
agent-browser set viewport 768 1024
agent-browser screenshot tablet.png
agent-browser set viewport 1280 720
agent-browser screenshot desktop.png
```

### Upload / dialogs / diff
```bash
agent-browser upload <selector> <file>
agent-browser dialog accept
agent-browser diff url <url-a> <url-b> --screenshot
```

## Output

- Tested path and scope
- Pass/fail observations
- Evidence paths (screenshots, command outputs)