Browser Harness

Use it when

• Your agent needs to act inside your real logged-in browser sessions instead of a clean headless profile that hits login walls
• You want browser automation that runs against local Chrome over one CDP websocket rather than being locked to a hosted cloud browser product
• Repeated site tasks keep re-discovering the same flows; you want the harness to persist agent-written helpers and domain skills across runs

What it solves

Not the fit when

• Scripted cross-browser end-to-end test suites
• Sites or accounts you are not authorized to automate
• Unsupervised consequential actions in a logged-in personal browser
• Stealth, proxy, CAPTCHA, sub-agent, or browser-fleet features unless using the separate optional Browser Use Cloud tier
• Hand-authoring domain skill files; the project asks that reusable skills be generated from successful harness runs
• scripted cross-browser e2e test suites
• html parsing without a browser
• mobile app automation

Install

Paste the README setup prompt into Claude Code or Codex: it installs browser-harness with uv on Python 3.12 and registers the skill via `browser-harness skill` (fallback guide: install.md in the repo)

Invoke

Enable the checkbox at chrome://inspect/#remote-debugging and allow the attach popup; the agent then runs `browser-harness` helper calls (e.g. print(page_info())) against your live Chrome, writing missing helpers into agent-workspace/agent_helpers.py as it works

Alternatives

No reviewed alternatives recorded yet.