browser-harness

Active
GitHub Python MIT

Description

Browser Harness | Self-healing harness that enables LLMs to complete any task.

Key Features

  • Thin CDP harness connecting LLMs directly to a real Chrome browser
  • Self-healing: agent writes missing helpers during execution
  • Domain skills system with community-contributed per-site playbooks
  • Works with Claude Code and Codex via simple setup prompt
  • Free cloud tier with stealth, sub-agents, and captcha solving
  • ~1k lines across 4 core files — minimal and extensible architecture

Use Cases

💡 Automating complex browser tasks with LLM-driven agents
💡 Web scraping and data extraction from authenticated sessions
💡 QA testing and dogfooding web applications end-to-end
💡 File uploads, form filling, and dialog handling via agent control
💡 Building reusable domain skills for recurring browser workflows

Tags

browser agent automation python testing

Categories

🌐 Browser Agent ⚡ Agent Tools

Quick Start

1. Paste the setup prompt into Claude Code or Codex to install browser-harness.
2. Open chrome://inspect/#remote-debugging and enable the remote debugging checkbox.
3. Allow the per-attach popup when prompted (Chrome 144+).
4. Start giving browser tasks to the agent — it will navigate, interact, and write helpers as needed.
Visit GitHub Visit Website

Related Projects