general-browsing

name	general-browsing
description	Browse websites like a human to avoid bot detection. Build browsing history before target site. Navigate by clicking elements instead of direct URLs. Auto-loaded when using browser tools.

Anti-Bot Browsing Patterns

Bot detection systems flag automated browsers by watching for: no browsing history, direct URL navigation with no referer, instant page loads with no human-like exploration, and skipping cookie/consent banners.

Rule 1: Build Browsing History First

Before navigating to your target site, visit 2-3 benign sites to create a realistic browsing trail. This establishes cookies, referer headers, and a history graph that looks human.

Warm-up Sequence

When starting a fresh browser session or visiting a target site for the first time:

1. Navigate to a search engine (rotate: google.com, bing.com, duckduckgo.com)
2. Take a snapshot, wait 2-3 seconds (simulate reading)
3. Navigate to an innocuous site (wikipedia.org, reddit.com, github.com)
4. Take a snapshot, scroll briefly
5. NOW navigate to your target site

Warm-up Site Rotation

Don't always use the same warm-up sites. Pick 2-3 from this list in random order:

• Search engines: google.com, bing.com, duckduckgo.com
• Reference: wikipedia.org, stackoverflow.com
• Social: reddit.com, news.ycombinator.com
• Developer: github.com, npmjs.com

When Warm-up is Required

• Always on first page load of a new session
• Always after container restart (new fingerprint)
• Always for sites known to use aggressive bot detection (Upwork, LinkedIn, etc.)
• Optional if you already have an active session with cookies and the target site already trusts you

Rule 2: Click, Don't Navigate

Once you're on the target site, never use browser_navigate to move between pages. Instead:

1. Take a browser_snapshot
2. Find the link/button that leads to your destination
3. Use browser_click on that element

Why

• browser_navigate to a URL has no referer context — bot-like
• Clicking a link sends the correct referer header
• Bot systems track navigation method, not just destination
• Clicking creates mouse movement and interaction patterns

Example: Login Page

# BAD - direct URL, no referer, bot-like
browser_navigate "https://www.target.com/login"

# GOOD - natural click-based navigation
browser_navigate "https://www.target.com/"        # initial load (with warm-up first)
browser_snapshot                                   # find the "Log In" link
browser_click "@eN"                                # click the login link/button
browser_snapshot                                   # now you're on the login page naturally

Example: Finding Jobs on Upwork

# BAD
browser_navigate "https://www.upwork.com/nx/search/jobs/?q=python"

# GOOD
browser_navigate "https://www.upwork.com/"        # homepage
browser_snapshot                                   # find search or navigation elements
browser_click "@eN"                                # click "Find Work" or search input
browser_fill "@eM" "python"                        # type in search (uses humanized typing)
browser_click "@eP"                                # click search button
browser_snapshot                                   # now you're on results page naturally

Rule 3: Handle Cookie/Consent Banners

Always dismiss cookie consent dialogs before proceeding. These are common on EU-targeted sites and are a tell if ignored.

browser_snapshot                                   # look for consent banner
browser_click "@eN"                                # click "Accept" or "Close"
browser_snapshot                                   # proceed with page (refs reset after dismiss!)

Rule 4: Add Human-Like Delays

Between actions, use brief pauses. Don't rapid-fire tool calls:

• After page load: wait for navigation or snapshot to confirm load
• Before typing: snapshot first to verify element is ready
• Between form fields: don't fill username and password in the same millisecond
• After clicking a button: wait for navigation/response

Rule 5: Handle Stale Refs

Refs (element references like @e2) invalidate after any navigation or page change. Always take a new browser_snapshot before interacting with elements — especially after:

• Navigation
• Clicking something that changes the page
• Scrolling (if lazy-loaded content appears)
• Dismissing dialogs/modals

# WRONG - refs from old snapshot
browser_snapshot          # refs: e1-e13
browser_click "@e3"       # page changes
browser_click "@e7"       # ERROR: @e7 is stale now

# RIGHT - refresh refs after page change
browser_snapshot          # refs: e1-e13
browser_click "@e3"       # page changes
browser_snapshot          # NEW refs: e1-e11
browser_click "@e5"       # use fresh refs

Complete Example: Logging into Upwork

# Step 1: Warm-up
browser_navigate "https://www.google.com"
browser_snapshot                    # browse briefly

browser_navigate "https://en.wikipedia.org"
browser_snapshot
browser_scroll "down"

# Step 2: Go to target homepage
browser_navigate "https://www.upwork.com/"
browser_wait_for_navigation         # handle Cloudflare if present
  targetUrl: "upwork.com"
  timeout: 15000
browser_snapshot

# Step 3: Dismiss cookie banner if present
# (click "Close" or "Accept" on the consent dialog)
browser_click "@e115"               # cookie banner close button

# Step 4: Navigate to login by CLICKING, not URL
browser_snapshot                    # fresh refs after banner dismiss
browser_click "@e9"                 # click the "Log In" link in nav
browser_snapshot                    # now on login page with proper referer

# Step 5: Fill credentials (humanized typing is automatic)
browser_click "@e2"                 # click username field
browser_fill "@e2" "[e-mail]"
browser_click "@e3"                 # click Continue or go to password
browser_snapshot                    # fresh refs
browser_fill "@ePasswordRef" "password123"

# Step 6: Submit
browser_snapshot                    # fresh refs
browser_click "@eSubmit"            # click Continue/Submit button
browser_wait_for_navigation
  targetUrl: "upwork.com"
  timeout: 15000