← Back to Skills Marketplace
andyless

Human Browser Use

by less · GitHub ↗ · v0.2.1 · MIT-0
cross-platform ⚠ suspicious
350
Downloads
0
Stars
3
Active Installs
1
Versions
Install in OpenClaw
/install human-browser-use
Description
Provides human-like browser automation with stealth features to bypass anti-bot measures using natural mouse, keyboard, and scroll behaviors.
README (SKILL.md)

human-browser-use Skill

Human-like browser automation extension for browser-use.

When to use

Use human-browser-use instead of raw browser-use when:

  • The target site has anti-bot detection (Cloudflare, reCAPTCHA, DataDome, etc.)
  • You need mouse movements to look like a real person
  • You need typing to have natural rhythm and occasional typos
  • You need to hide automation fingerprints (navigator.webdriver, WebGL, etc.)

Installation

pip install human-browser-use

CLI (preferred for quick tasks)

hbu open https://example.com       # Navigate
hbu state                           # See elements
hbu click 5                         # Click (human-like trajectory)
hbu type "Hello"                    # Type (human-like dynamics)
hbu screenshot page.png             # Screenshot
hbu close                           # Close

All browser-use CLI commands work with hbu. The browser stays alive between commands.

Python API

import asyncio
from human_browser_use import HumanBrowserSession, HumanBrowserProfile, HumanBehaviorConfig

async def main():
    session = HumanBrowserSession(
        human_config=HumanBehaviorConfig(),
        browser_profile=HumanBrowserProfile(headless=False),
    )
    await session.start()
    await session.navigate_to("https://example.com")

    page = await session.get_current_page()           # Returns HumanPage
    els = await page.get_elements_by_css_selector("input")  # Returns HumanElement[]
    await els[0].click()                               # Human-like Bezier trajectory
    await els[0].fill("hello world")                   # Human-like typing dynamics
    await page.press("Enter")

    await session.reset()

asyncio.run(main())

With browser-use Agent

from browser_use import Agent
from langchain_openai import ChatOpenAI
from human_browser_use import HumanBrowserSession, HumanBrowserProfile, HumanBehaviorConfig

agent = Agent(
    task="Your task here",
    llm=ChatOpenAI(model="gpt-4o"),
    browser_session=HumanBrowserSession(
        human_config=HumanBehaviorConfig(),
        browser_profile=HumanBrowserProfile(headless=False),
    ),
)
await agent.run()

API reference

Class Replaces Purpose
HumanBrowserSession BrowserSession Session with human behavior + stealth JS
HumanBrowserProfile BrowserProfile Chrome flags to hide automation fingerprints
HumanBehaviorConfig Master config (mouse, keyboard, scroll, timing)

HumanBrowserSession

session = HumanBrowserSession(human_config=config, browser_profile=profile)
await session.start()
await session.navigate_to(url)
page = await session.get_current_page()   # HumanPage
pages = await session.get_pages()         # list[HumanPage]
await session.reset()

HumanPage (returned by session)

elements = await page.get_elements_by_css_selector("selector")  # list[HumanElement]
element = await page.get_element("selector")                     # HumanElement | None
await page.press("Enter")
await page.goto("https://...")

HumanElement (returned by page)

await element.click()                    # Bezier trajectory + variable press duration
await element.fill("text")              # Lognormal delays + typo simulation
await element.fill("text", clear=False) # Append without clearing

Configuration cheatsheet

config = HumanBehaviorConfig()

# Mouse
config.mouse.overshoot_probability = 0.15    # Overshoot chance
config.mouse.click_offset_sigma = 3.0        # Click position randomness (px)
config.mouse.press_duration_range = (0.05, 0.15)

# Keyboard
config.keyboard.delay_mu = 4.17              # Lognormal mean → ~65ms
config.keyboard.typo_probability = 0.02      # Typo chance per key
config.keyboard.common_bigram_factor = 0.7   # "th","er" 30% faster

# Scroll
config.scroll.impulse_delta_range = (80, 200)
config.scroll.inertia_decay = 0.85

# Timing
config.timing.pre_action_delay_range = (0.1, 0.3)

# Feature toggles
config.enable_stealth = True
config.enable_human_mouse = True
config.enable_human_keyboard = True
config.enable_human_scroll = True

Important rules

  • Always import from human_browser_use, not browser_use
  • Use HumanBrowserSession + HumanBrowserProfile (not BrowserSession / BrowserProfile)
  • Get elements via page.get_elements_by_css_selector() — they return HumanElement with human-like behavior
  • Do NOT use base Element class directly — it bypasses human behavior
  • If using a local proxy, set os.environ['no_proxy'] = 'localhost,127.0.0.1' before creating the session
Usage Guidance
This skill's behavior broadly matches its description, but it has notable gaps: it instructs you to pip-install an unverified package (no source/homepage provided) and does not declare required runtime dependencies (chromium/Chrome, Playwright/Selenium, or the underlying browser-use package). Before installing, ask the publisher for: (1) the package source (PyPI link and/or GitHub repo) and a checksum or vouches for the release; (2) a complete dependency list and instructions for required browser binaries; (3) an explicit statement about intended/legal use (evading anti-bot protections can violate terms of service and laws). If you proceed, review the package contents (or install inside an isolated VM/container), avoid using real credentials or production accounts with it, and prefer packages with a verifiable repository and signed releases.
Capability Analysis
Type: OpenClaw Skill Name: human-browser-use Version: 0.2.1 The `human-browser-use` skill provides browser automation features specifically designed to evade anti-bot protections (e.g., Cloudflare, reCAPTCHA) and hide automation fingerprints, as described in `skill.md`. These capabilities are high-risk as they facilitate the bypass of web security controls. The documentation also instructs the AI agent to modify environment variables (`no_proxy`) and strictly use specialized wrappers to maintain stealth. While no explicit malicious intent like data exfiltration was found, the inherent risk of automated security evasion and a future-dated timestamp in `_meta.json` (2026) warrant a suspicious classification. IOC: `pip install human-browser-use`.
Capability Assessment
Purpose & Capability
The SKILL.md content aligns with the stated purpose (human-like, stealthy browser automation). However, the skill instructs use of a pip package that is not provided in the registry (no source/homepage) and does not declare required runtime dependencies (browser binary, Playwright/Selenium, or browser-use package). That gap is an incoherence: a browser-automation skill normally needs to declare external runtime dependencies.
Instruction Scope
Instructions are focused on browser automation and explicitly emphasize evading anti-bot measures (stealth JS, hiding navigator.webdriver, etc.). They do not ask the agent to read unrelated files or secrets. The explicit goal of bypassing anti-bot protections is legitimate for testing but also has high potential for misuse; this is a policy/ethical concern rather than a direct technical incoherence.
Install Mechanism
There is no install spec in the registry and no code files. The SKILL.md tells users to run `pip install human-browser-use`, but the package source, checksum, or repository are not provided and the registry lists no install mechanism. Telling users to install an unverified PyPI package from an unknown source is risky and inconsistent with the absence of an install entry.
Credentials
The skill does not request environment variables, secrets, or config paths. The only env-related guidance is setting no_proxy for local proxies, which is proportional and not suspicious.
Persistence & Privilege
The skill is not always-enabled and does not request elevated persistence. It does not claim to modify other skills or global agent settings.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install human-browser-use
  3. After installation, invoke the skill by name or use /human-browser-use
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v0.2.1
- Added documentation for human-browser-use, detailing usage with CLI, Python, and agent patterns. - Outlines API replacements: HumanBrowserSession, HumanBrowserProfile, and HumanBehaviorConfig for superior stealth and human-like interaction. - Provides configuration details for customizing mouse, keyboard, scroll, and timing behaviors to mimic real users. - Emphasizes best practices: always import from human_browser_use and use HumanElement for actions to ensure human-like automation. - Includes installation, usage examples, and a configuration cheatsheet for quick reference.
Metadata
Slug human-browser-use
Version 0.2.1
License MIT-0
All-time Installs 3
Active Installs 3
Total Versions 1
Frequently Asked Questions

What is Human Browser Use?

Provides human-like browser automation with stealth features to bypass anti-bot measures using natural mouse, keyboard, and scroll behaviors. It is an AI Agent Skill for Claude Code / OpenClaw, with 350 downloads so far.

How do I install Human Browser Use?

Run "/install human-browser-use" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is Human Browser Use free?

Yes, Human Browser Use is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does Human Browser Use support?

Human Browser Use is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created Human Browser Use?

It is built and maintained by less (@andyless); the current version is v0.2.1.

💬 Comments