skills

stealth-cli

Anti-detection browser CLI powered by Camoufox. Browse the web, search, extract data, take screenshots, and crawl pages — all with C++ level fingerprint spoofing that bypasses Cloudflare, Google, and most bot detection systems.

When to use

• User needs to browse a website that blocks automated tools
• User needs to scrape data from protected pages
• User needs to search Google/Bing/DuckDuckGo without being blocked
• User needs screenshots of pages behind anti-bot protection
• User needs to monitor a page for changes
• User needs to extract structured data (links, images, meta tags)
• User needs to crawl a site with anti-detection
• User needs a human to complete login, CAPTCHA, 2FA, or OAuth in a persistent profile

Prerequisites

cd ~/Desktop/stealth-cli
npm install    # installs deps + downloads Camoufox (~300MB first time)

Commands

Browse a page

# Text output (default)
stealth browse https://example.com

# JSON output with metadata
stealth browse https://example.com -f json

# With proxy
stealth browse https://example.com --proxy http://user:pass@host:port

# Create a named profile from a preset once, then reuse it
stealth profile create work --preset us-desktop
stealth browse https://example.com --profile work

Human-assisted authentication

# The browser is headed, ignores stdin, and stays alive until all windows close.
# Cookies are checkpointed to the profile every second.
stealth profile create work --preset us-laptop
stealth open https://example.com/login --profile work

# Persist a named session as well as the profile
stealth open --url https://example.com/login --profile work --session login-flow

Use open, not browse --no-headless, for CAPTCHA, 2FA, OAuth consent, or any flow where a human needs time to interact. open always bypasses the daemon. If profile and session are combined, profile cookies are canonical and a session linked to another profile is rejected. If the browser process terminates before the final live save, the latest durable checkpoint is retained. --checkpoint-interval accepts integer values from 250 through 60000 ms (default: 1000).

Screenshot

stealth screenshot https://example.com -o page.png
stealth screenshot https://example.com --full -o full.png    # full page
stealth screenshot https://example.com --width 1920 --height 1080

Search (bypasses anti-bot)

# Google (auto-humanized: visits homepage, types query, presses enter)
stealth search google "web scraping tools" -f json

# Other engines
stealth search duckduckgo "query" -f json
stealth search bing "query" -f json
stealth search youtube "query" -f json
stealth search github "query" -f json

# With warmup (visit random site first to build browsing history)
stealth search google "query" -f json --warmup

# Supported: google, duckduckgo, bing, youtube, github, amazon,
#   reddit, wikipedia, twitter, linkedin, tiktok, stackoverflow,
#   npmjs, yelp

Extract data

# Extract all links
stealth extract https://example.com --links

# Extract meta tags (title, description, og:*)
stealth extract https://example.com --meta

# Extract all headings
stealth extract https://example.com --headers

# Extract by CSS selector
stealth extract https://example.com -s ".product-title" --all

# Extract image URLs
stealth extract https://example.com --images

Crawl

# Crawl 2 levels deep, max 20 pages
stealth crawl https://example.com -d 2 -l 20

# Save to file
stealth crawl https://example.com -o results.jsonl

# Filter URLs
stealth crawl https://example.com --include "blog" --exclude "login"

# With delay between pages
stealth crawl https://example.com --delay 2000

Monitor for changes

# Watch for any change (check every 60s)
stealth monitor https://example.com/price -s ".price" -i 60

# Alert when text appears
stealth monitor https://example.com --contains "In Stock"

# Alert when text disappears
stealth monitor https://example.com --not-contains "Sold Out"

# JSON output for piping
stealth monitor https://example.com --json -n 10

Batch processing

# Create a URL list file
echo "https://example.com\nhttps://httpbin.org/ip" > urls.txt

# Browse all URLs
stealth batch urls.txt -c browse --skip-errors

# Screenshot all URLs
stealth batch urls.txt -c screenshot -o ./screenshots/

Browser fingerprint check

# Show current fingerprint
stealth fingerprint

# Run anti-detection tests
stealth fingerprint --check

# Compare fingerprints across multiple launches
stealth fingerprint --compare 3

Profile management (persistent identities)

# Create from preset
stealth profile create work --preset us-desktop
stealth profile create jp --preset jp-desktop

# Create random
stealth profile create random1 --random

# List profiles
stealth profile list

# One-shot command (auto-saves cookies before it closes)
stealth browse https://example.com --profile work

# Human login flow (periodic checkpoints + final save on window close)
stealth open https://example.com/login --profile work

# Available presets: us-desktop, us-laptop, uk-desktop, de-desktop,
#   jp-desktop, cn-desktop, mobile-ios, mobile-android

Session persistence

# Link a named session to the existing work profile
stealth browse https://example.com --session my-task --profile work

# Session-only open restores the linked profile and saved URL
stealth open --session my-task

# An explicit initial URL skips the saved URL
stealth open https://other.com --session my-task

A session linked to a profile automatically restores that profile when only --session is supplied. For open and interactive, an explicit initial URL skips the session's saved URL before navigation. Profile and session names accept only letters, numbers, underscores, and hyphens, use lowercase canonical identities, and reject Windows device basenames such as CON, NUL, COM1, and LPT1 on every platform. Older versions sanitized unsupported filename characters to underscores; reuse that basename (for example, work.prod became work_prod). stealth profile list shows profile basenames. There is no session-list command, so inspect $STEALTH_HOME/sessions (default ~/.stealth/sessions) for a legacy session filename. Compatible legacy metadata is rewritten on the next successful save, but files are not automatically renamed. If a linked profile is missing or stored state is malformed, startup fails before browser launch instead of silently using another identity.

Named profile and session browser state is single-writer: browser lifetimes and standalone mutations use the same lease protocol. Concurrent reuse fails. Crash-left locks are not auto-removed; verify that no process owns the state before removing the exact lock path printed by the CLI. STEALTH_HOME relocates profiles, sessions, and their locks from the default ~/.stealth; its state paths must not be symlinks. POSIX systems enforce owner-only directory/file modes (0700/0600), including config and proxy-pool credential storage; Windows skips POSIX mode-bit enforcement, so use user-only ACLs and avoid shared directories. Config, proxy-pool, and daemon paths still use ~/.stealth in the current source.

Proxy pool

stealth proxy add http://user:pass@host:port --label us-east
stealth proxy list
stealth proxy test          # test all proxies

# Auto-rotate through pool
stealth browse https://example.com --proxy-rotate
stealth crawl https://example.com --proxy-rotate

Daemon mode (fast startup)

stealth daemon start        # launch background browser
stealth browse https://...  # instant (~1s vs ~6s cold start)
stealth daemon status
stealth daemon stop

HTTP API server

stealth serve --port 9377

# Then call via HTTP:
curl -X POST http://localhost:9377/tabs -d '{"url":"https://example.com"}'
curl http://localhost:9377/tabs/tab-1/text
curl http://localhost:9377/tabs/tab-1/screenshot

Interactive REPL

stealth interactive --url https://example.com

# Commands: goto, search, click, hclick, type, htype, scroll,
#   snapshot, text, title, url, links, screenshot, back, forward,
#   reload, eval, wait, help, exit

Output format

Commands that produce page or search data may support --format or -f (check stealth <command> --help):

• text — human-readable (default)
• json — structured JSON
• jsonl — one JSON object per line
• snapshot — accessibility tree snapshot
• markdown — markdown formatted

stdout contains data, stderr contains status messages. Pipe-friendly:

stealth search google "query" -f json | jq '.results[].url'
stealth extract https://example.com --links -f json | jq '.[].url'

Common browser-command options

Option availability varies by command; run stealth <command> --help for the exact set.