camoufox-stealth-browser

name	camoufox-stealth-browser
homepage	https://github.com/kesslerio/camoufox-stealth-browser-clawhub-skill
description	Explicit-only stealth browser automation with Camoufox for hostile sites that block native OpenClaw browser, standard Playwright, or Selenium flows. Default to the native OpenClaw browser for ordinary browsing, booking flows, calendars, forms, screenshots, and generic site testing. Use Camoufox only when the user explicitly asks for Camoufox, a stealth browser, bot bypass, anti-bot evasion, Cloudflare/Datadome bypass, persistent login/session reuse, or a target has already blocked the native browser lane. Browser lane only; API helpers are secondary.
metadata	{"openclaw":{"emoji":"🦊","requires":{"bins":[],"env":[]}}}

Camoufox Stealth Browser 🦊

Default browser lane: native OpenClaw browser.

Do not use this skill for ordinary browser automation. Use native OpenClaw browser first for normal browsing, appointment booking, dynamic calendars, forms, screenshots, accessibility snapshots, and generic site testing.

Use Camoufox only when one of these is true:

• The user explicitly asks for Camoufox.
• The user explicitly asks for a stealth browser, bot bypass, anti-bot evasion, Cloudflare bypass, Datadome bypass, or similar.
• Native OpenClaw browser already reached the site and was blocked by anti-bot or fingerprinting defenses.
• The task requires a persistent hostile-site stealth session and the user accepts that tradeoff.

Camoufox is the hostile-site browser lane, not the default browser lane. If native OpenClaw browser is unavailable, report that directly instead of silently switching to Camoufox unless the user asked for stealth/bypass behavior.

Why Camoufox

Approach	Patch level	Typical weakness
Camoufox	Browser/runtime	Harder to fingerprint with timing and JS consistency checks
undetected-chromedriver	JS/runtime glue	Timing and environment mismatches
puppeteer-stealth	JS injection	Patches land after page startup
playwright-stealth	JS injection	Same class of weakness

Runtime Selection

For explicit Camoufox workflows, the skill uses this order:

1. OpenClaw/AlphaClaw bridge at /data/bin/camoufox-nixos, when present
2. camoufox-nixos from PATH
3. distrobox with pybox
4. clear setup failure if neither exists

Inside OpenClaw/AlphaClaw, prefer /data/bin/camoufox-nixos. Do not override the scripts to use /run/current-system/sw/bin/camoufox-nixos unless you are deliberately diagnosing the direct host wrapper; in this environment that path can report misleading missing-venv errors such as /data/.local/venvs/camoufox/bin/python.

The repo still carries a separate curl_cffi helper, but it is not the primary routing target for this skill.

Quick Start

Browser workflows

The browser scripts self-detect runtime. Use them directly:

./scripts/camoufox-fetch.py "https://example.com" --headless

./scripts/camoufox-session.py \
  --profile economist \
  --status "https://www.economist.com"

Do not frame browser execution as "importing Camoufox into system Python" or "switching to the proper runtime." Run the wrapper scripts directly and let them choose the runtime.

Runtime Failure Reporting

When browser execution fails, report the observed runtime failure first.

• If camoufox-nixos times out or exits before navigation, say it failed before page navigation and treat it as a host runtime launch issue until proven otherwise.
• Do not turn a wrapper startup failure into a target-site blocking diagnosis.
• Do not claim a "legacy Python path worked" unless you actually ran that exact path and verified success.
• If you suggest the distrobox fallback, present it as an alternate documented browser lane, not proof that system Python imports are the real problem.

Optional fallback/API setup

If camoufox-nixos is missing, or if you need the curl_cffi lane, run:

bash scripts/setup.sh

That script configures the distrobox fallback when pybox is available and tells you what is missing when it is not.

If you just patched the host-local NixOS wrapper itself, remember that the live camoufox-nixos command does not change until the host runs:

sudo nixos-rebuild switch --flake /etc/nixos#nixos

When To Use

• Standard Playwright or Selenium gets blocked
• The site shows Cloudflare challenge loops
• You need persistent authenticated browsing for hostile or paywalled sites
• You need actual stealth rather than generic browser automation
• You need login/session reuse on a host that already has camoufox-nixos

Do not use this skill for ordinary browsing or generic site testing. Use native OpenClaw browser for that.

Workflow Summary

Protected-page fetch

./scripts/camoufox-fetch.py \
  "https://www.yelp.com/biz/example" \
  --headless \
  --wait 8 \
  --screenshot yelp.png \
  --output yelp.html

Persistent session

# Interactive login
./scripts/camoufox-session.py \
  --profile airbnb \
  --login "https://www.airbnb.com/account-settings"

# Reuse saved session
./scripts/camoufox-session.py \
  --profile airbnb \
  --headless "https://www.airbnb.com/trips"

# Check session status
./scripts/camoufox-session.py \
  --profile airbnb \
  --status "https://www.airbnb.com"

State Ownership

This skill owns no durable state of its own. Browser profile state belongs to the selected runtime:

• OpenClaw/AlphaClaw bridge (/data/bin/camoufox-nixos): bridge-managed host state mapping
• Host-native (camoufox-nixos): ~/.cache/camoufox-nixos
• Legacy distrobox fallback: ~/.stealth-browser/profiles/<name>/

Implications:

• Persistent session reuse still works in both lanes because the runtimes own their own profile/cache locations.
• --import-cookies is a legacy fallback feature. If you ask for it without the distrobox lane, the script fails clearly instead of pretending parity.
• --export-cookies continues to work.

Gotchas

See references/gotchas.md for the non-obvious footguns: runtime-framing mistakes, browser lane vs API helper confusion, headed-login expectations, Linux-only fallback assumptions, and host-native state-path differences.

Secondary API Helper

curl_cffi remains in this repo as a secondary API-only helper. It is useful when:

• there is no browser interaction requirement
• you already know the API endpoint
• browser overhead would be wasted

It is not the primary skill contract. Current repo guidance for it is still the legacy distrobox path:

distrobox enter pybox -- python3.14 scripts/curl-api.py "https://api.example.com"

Non-NixOS And Missing Runtime

• NixOS hosts with camoufox-nixos: browser lane is host-native by default
• Other Linux hosts: use the distrobox fallback
• macOS / Windows: camoufox-nixos is not the portability story; the repo’s portable browser guidance is still the distrobox fallback where available

This skill does not try to teach every machine how to recreate camoufox-nixos. That wrapper is host-specific, and the distrobox lane is the compatibility path where available.

Proxy Reminder

For Airbnb, Yelp, Datadome, and similar targets:

• datacenter IPs often get blocked immediately
• residential or mobile proxies are usually required
• sticky sessions matter more than rotating every request

See references/proxy-setup.md.

Remote / Headed Login Notes

Interactive login still needs a visible browser window regardless of runtime. If you are remote, use a display-capable setup such as:

• local desktop session
• SSH with display forwarding where supported
• VNC or similar remote desktop

On TTY-only or headless shell sessions, headed camoufox-nixos open should fail fast with display_missing. That is expected. Use --headless unless you intentionally attached a real display.

Troubleshooting

Problem	Meaning	What to do
No supported browser runtime found	Neither camoufox-nixos nor valid distrobox fallback was detected	Install the host wrapper or configure distrobox plus pybox
display_missing	You tried a headed host-native launch on a session with no DISPLAY or WAYLAND_DISPLAY	This is expected on TTY-only hosts and remote shells; rerun with --headless or use a real graphical session
camoufox-nixos times out before navigation	Browser wrapper failed to launch cleanly on the host	Report it as a host runtime launch issue first; only discuss site blocking after the browser actually reaches the target
--import-cookies requires the legacy distrobox fallback	Host-native lane cannot honestly reproduce that legacy import flow	Use the fallback lane for that operation
Browser lane works but curl-api.py does not	curl_cffi lane is still legacy-path setup in this repo	Run bash scripts/setup.sh
Immediate block or challenge loop	Proxy quality or behavior issue	Use residential/mobile proxy and increase wait time

References

• README.md — repo overview
• references/gotchas.md — recurring footguns and local assumptions
• references/proxy-setup.md — proxy guidance
• references/fingerprint-checks.md — anti-bot fingerprint categories