Ghost Browser

Automated Chrome browser using nodriver for AI agent web tasks. Full CLI control with LLM-optimized commands — text-based interaction, markdown output, sessi...

This skill mostly does what it says (Chrome automation using nodriver), but there are packaging and install inconsistencies and it will store session cookies and profiles locally. Before installing: 1) Review the full Python script (scripts/stealth_browser.py) for any network endpoints or unexpected behavior you don't trust. 2) Do not run setup.sh blindly — it installs packages and attempts to make a CLI executable but references a non-existent path (scripts/ghost-browser), indicating a broken package. 3) Consider running the skill in an isolated environment (VM/container) or test account to avoid exposing real credentials. 4) If you plan to persist sessions, be aware cookies/localStorage are stored under ~/.openclaw/... and can contain sensitive tokens. 5) Prefer skills with an identifiable source/homepage or official releases; if you need this capability, request a corrected package (matching filenames and a clear install spec) or a signed release from the author.

Type: OpenClaw Skill Name: neo-stealth-browser Version: 1.0.4 The OpenClaw AgentSkills skill bundle 'neo-stealth-browser' is classified as suspicious due to its powerful capabilities that, while intended for legitimate browser automation and anti-bot evasion, pose significant risks if misused. Key indicators include the `eval` command in `scripts/stealth_browser.py` and `SKILL.md`, allowing arbitrary JavaScript execution within the browser context. Additionally, the skill supports downloading files from arbitrary URLs (`_handle_download` in `stealth_browser.py`) and installing/loading Chrome extensions (`cmd_install_extension`, `_handle_load_extension`), which could be exploited to fetch and execute malicious payloads or install compromised browser extensions. While the skill's documentation and code include defensive prompt injection and transparency regarding data storage, these powerful primitives, combined with advanced anti-detection techniques (e.g., mouse event patching, human-like clicks, reCAPTCHA evasion), elevate the risk profile to suspicious, as they could facilitate sophisticated attacks if the agent is compromised or given malicious instructions.