Browser Harness ♞

Connect an LLM directly to your real browser with a thin, editable CDP harness. For browser tasks where you need complete freedom.

One websocket to Chrome, nothing between. The agent writes what's missing during execution. The harness improves itself every run.

  ● agent: wants to upload a file
  │
  ● agent-workspace/agent_helpers.py → helper missing
  │
  ● agent writes it                         agent_helpers.py
  │                                                       + custom helper
  ✓ file uploaded

You will never use the browser again.

Install (v0.5.1+)

This fork is published to PyPI as browser-harness-extended — the original browser-harness name is reserved on PyPI. The Python import name is unchanged (import browser_harness still works).

pip install browser-harness-extended
# or via the GitHub tag (no PyPI required):
pip install "git+https://github.com/Abhishekchhetri020/browser-harness@v0.5.1"

What's new in v0.4.0 — per-user shadow cache (autotrace records, replay_skill deterministically replays at zero LLM cost), auto-codification of recurring flows into committable .py skills after 3 successful replays, bh-policy.toml trust layer with per-host glob allowlists, secret-scanned skill registry pipeline (publish_skill dry-run + gh-CLI live), multi-Chromium fan-out (launch_browser for Brave/Edge/Arc/Chromium variants), and an EU-friendly BH_LOCAL_ONLY=1 mode that asserts no harness-initiated network egress. Full notes in CHANGELOG.md. Public data-flow contract in docs/data-flow.md. Trust-layer schema in bh-policy.toml.example.

Setup prompt

Paste into Claude Code or Codex:

Set up https://github.com/browser-use/browser-harness for me.

Read `install.md` and follow the steps to install browser-harness and connect it to my browser.

The agent will open chrome://inspect/#remote-debugging. Tick the checkbox so the agent can connect to your browser:

Click Allow when the per-attach popup appears (Chrome 144+):

See agent-workspace/domain-skills/ for example tasks.

Free Browser Use Cloud browsers

Stealth, sub-agents, or headless deployment.
Browser Use Cloud free tier: 3 concurrent browsers, proxies, captcha solving, and more. No card required.

• Grab a key at cloud.browser-use.com/new-api-key
• Or let the agent sign up itself via docs.browser-use.com/llms.txt (setup flow + challenge context included).

Architecture (v0.4.0 — ~3.5k lines across the core package)

File	Purpose
src/browser_harness/helpers.py	Primitive surface auto-imported into -c scripts
src/browser_harness/daemon.py	Long-lived CDP holder + 500-event ring buffer, one per BU_NAME
src/browser_harness/admin.py	Daemon lifecycle, doctor, Browser-Use Cloud session admin
src/browser_harness/snapshot.py	AX-tree snapshot → r1/r2/... refs
src/browser_harness/observation.py	Network + console event drain
src/browser_harness/_audit.py	NDJSON audit log at ~/.browser-harness/audit.log
src/browser_harness/_trace.py	Per-user shadow cache: autotrace, replay_skill, list_traces
src/browser_harness/_codify.py	Auto-promote N-times-replayed traces → committable .py skills
src/browser_harness/_policy.py	bh-policy.toml trust layer; per-host glob allowlists
src/browser_harness/_skills.py	Skill registry: list / validate / package / publish
src/browser_harness/_browsers.py	Multi-Chromium detect / find_free_port / launch_browser
src/browser_harness/_egress.py	BH_LOCAL_ONLY egress gate (EU-friendly mode)
src/browser_harness/mcp_server.py	MCP tool surface for clients that prefer JSON-RPC over -c
src/browser_harness/run.py	The browser-harness CLI
src/browser_harness/_ipc.py	Token-authed unix socket / TCP loopback
agent-workspace/agent_helpers.py	Helper code the agent edits
agent-workspace/domain-skills/<host>/	Per-host skill folders (markdown + auto-codified .py)
bh-policy.toml.example	Trust-layer schema you can copy and edit
docs/data-flow.md	Public data-flow contract — what the harness reads, where it goes
docs/registry-template/	Drop-in template for standing up a community skill registry repo

179 unit tests. MIT (inherited from upstream browser-use/browser-harness).

Contributing

PRs and improvements welcome. The best way to help: contribute a new domain skill under agent-workspace/domain-skills/ for a site or task you use often (LinkedIn outreach, ordering on Amazon, filing expenses, etc.). Each skill teaches the agent the selectors, flows, and edge cases it would otherwise have to rediscover.

• Skills are written by the harness, not by you. Just run your task with the agent — when it figures something non-obvious out, it files the skill itself (see SKILL.md). Please don't hand-author skill files; agent-generated ones reflect what actually works in the browser.
• Open a PR with the generated agent-workspace/domain-skills/<site>/ folder — small and focused is great.
• Bug fixes, docs tweaks, and helper improvements are equally welcome.
• Browse existing skills (github/, linkedin/, amazon/, ...) to see the shape.

If you're not sure where to start, open an issue and we'll point you somewhere useful.

Domain skills

Set BH_DOMAIN_SKILLS=1 to enable agent-workspace/domain-skills/ — community-contributed per-site playbooks goto_url surfaces by domain. Contribute via PR.

---

The Bitter Lesson of Agent Harnesses · Web Agents That Actually Learn