browser-harness

name	browser-harness
description	Always use Browser Harness for web interaction that needs robust automation, downloads, uploads, screenshots, testing, or the user's live Chrome session.
dependencies	{"python":[],"node":[],"binaries":["browser-harness"]}

Browser Harness

Browser Harness controls Chrome directly through CDP. The default upstream checkout is at:

$env:USERPROFILE\Developer\browser-harness

The installed CLI is the normal entry point. The checkout is available for its current SKILL.md, install.md, interaction guides, and source diagnostics.

Usage from PowerShell

Pass Python helpers on standard input; -c and --setup are not valid commands:

@'
new_tab("https://example.com")
wait_for_load()
print(page_info())
'@ | browser-harness

Helpers are pre-imported. The daemon starts automatically on the first command. Use new_tab(url) for the first navigation and call wait_for_load() afterward.

Health and connection

browser-harness --doctor
browser-harness --reload
browser-harness recordings

If local Chrome cannot attach, open chrome://inspect/#remote-debugging, ask the user to enable Allow remote debugging for this browser instance, and wait for the user to approve Chrome's popup. Do not claim Browser Harness is unavailable until browser-harness --doctor or an actual harness command reports the failure.

Downloads, screenshots, and advanced interaction

Before inventing mechanics, read the relevant upstream guide under:

$env:USERPROFILE\Developer\browser-harness\interaction-skills\

Important guides include downloads.md, screenshots.md, uploads.md, tabs.md, dialogs.md, and print-as-pdf.md.

For a site-specific flow, inspect the official examples under $env:USERPROFILE\Developer\browser-harness\agent-workspace\domain-skills\. Task-specific helpers belong in Browser Harness's configured agent workspace, not in LimeBot core.

Recording privacy

Recordings are disabled by default. Preserve the user's configured preference. Never enable background recordings without explicit consent.