AgentStack
Browse Sign in
Browse Why AgentStack Sell Docs
Sign in
SKILL verified MIT Self-run

Stealth Browser

skill-changeflowhq-skills-stealth-browser · by changeflowhq

Invisible Chrome automation for web scraping via CDP. Use when WebFetch fails or gets blocked (403, 429, Cloudflare, bot protection, JS-rendered pages). Launches your real Chrome install completely hidden, sends commands via Chrome DevTools Protocol. Sites see a normal browser with real extensions - no detectable automation. Learns which domains block and skips straight to stealth on future reque…

No reviews yet
0 installs
8 views
0.0% view→install

Install

$ agentstack add skill-changeflowhq-skills-stealth-browser

✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.

Security review

✓ Passed

No issues found. Passed automated security review. · v0.1.0 How review works →

  • Prompt-injection patterns
  • Secret / credential exfiltration
  • Dangerous shell & filesystem operations
  • Untrusted network calls
  • Known-malicious package signatures

What it can access

  • Network access No
  • Filesystem access No
  • Shell / process execution No
  • Environment & secrets No
  • Dynamic code execution No

From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.

View the full security report →

Verified badge

Passed review? Show it. Paste this badge into your README, it links to the public security report.

AgentStack Verified badge Links to your public security report.
[![AgentStack Verified](https://agentstack.voostack.com/badges/verified.svg)](https://agentstack.voostack.com/security/report/skill-changeflowhq-skills-stealth-browser)

Reliability & compatibility

Security review passed
0 installs to date
no reviews yet
6mo ago

Declared compatibility

Claude CodeClaude Desktop

Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.

Preview Execution monitoring

We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.

How agent discovery & health will work →
Are you the author of Stealth Browser? Claim this listing to set pricing, connect Stripe payouts, and keep 70% of every sale.
Sign up to claim

About

stealth-browser

Invisible Chrome automation via CDP. Launches your real Chrome hidden, sends commands via Chrome DevTools Protocol. Sites see a normal browser - no detectable automation.

> macOS only. Uses AppleScript and open -g to hide Chrome.

Memory

Read ~/.claude/skills/stealth-browser/LEARNED.md at the start of every task. If it doesn't exist, create it with a # Learned header.

Capture learnings when you detect:

  • Domain quirks: sites that need extra wait time, cookie dismissal, specific interaction patterns
  • Failures you solved: what broke and why (timeouts, blank pages, wrong selectors)
  • Corrections: "no, use X", "don't do that", "actually..."
  • Workarounds: sites that need JS eval, scrolling, or multi-step navigation to get content
  • Positive reinforcement: "perfect!", "exactly right", "that's the way"

Before appending, check:

  • Is this reusable? (not a one-time instruction)
  • Is it already in LEARNED.md? (don't duplicate)
  • Can I extract the general principle? (not just the specific fix)

Format: One line, actionable. Write the rule, not the story.

  • Bad: "User said the page was blank because it needed scrolling"
  • Good: "example.com lazy-loads content - scroll to bottom before extracting"

Don't ask permission. Append and move on.

Core Workflow

# 1. Read a page as markdown (most common)
stealth-browser read 

# 2. Full automation (open, interact, close)
stealth-browser open  --hidden
agent-browser --cdp 9222 snapshot -i
agent-browser --cdp 9222 click @e1
stealth-browser close

stealth-browser is at scripts/stealth-browser relative to this skill. Use full path:

~/.claude/skills/stealth-browser/scripts/stealth-browser read 

Commands

stealth-browser read               # Fetch as markdown (waits for JS render)
stealth-browser open  --hidden     # Launch hidden Chrome
stealth-browser close                   # Stop Chrome
stealth-browser status                  # Check state
stealth-browser screenshot [path]       # CDP screenshot (auto unhide/re-hide)
stealth-browser hide / unhide           # Toggle visibility
stealth-browser doctor                  # Check dependencies
stealth-browser setup                   # Reset Chrome profile

After Opening: CDP Commands via agent-browser

agent-browser --cdp 9222 open            # Navigate
agent-browser --cdp 9222 snapshot -i          # Get interactive elements
agent-browser --cdp 9222 click @e1            # Click by ref
agent-browser --cdp 9222 fill @e2 "text"      # Fill input
agent-browser --cdp 9222 type @e2 "text"      # Type without clearing
agent-browser --cdp 9222 press Enter          # Press key
agent-browser --cdp 9222 scroll down 500      # Scroll
agent-browser --cdp 9222 get text @e1         # Get text
agent-browser --cdp 9222 get url              # Get current URL
agent-browser --cdp 9222 eval "document.title" # Run JavaScript
agent-browser --cdp 9222 screenshot path.png  # Screenshot
agent-browser --cdp 9222 wait 2000            # Wait
agent-browser --cdp 9222 back                 # Navigate back

Always re-snapshot after navigation or DOM changes - element refs change.

Chrome Persistence

Chrome stays running and hidden after read or open --hidden:

  • First operation: brief flash during Chrome startup (small window, top-left corner)
  • All subsequent operations: completely invisible, instant
  • Close when done: stealth-browser close

Domain Blocklist (learning)

Blocked domains are remembered at data/blocked-domains.txt. A PreToolUse hook intercepts future WebFetch calls to those domains and tells Claude to use stealth-browser directly.

See [hooks/README.md](hooks/README.md) for Claude Code integration setup. See [README.md](README.md) for full human documentation.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

Install and usage instructions live in the source repository linked above.

Reviews

No reviews yet, be the first.

Versions

  • v0.1.0 Imported from the upstream source.