← Back to Skills Marketplace
Vanzhangsh Skills
by
vanzhangsh
· GitHub ↗
· v1.0.0
402
Downloads
0
Stars
0
Active Installs
1
Versions
Install in OpenClaw
/install vanzhangsh-skills
Description
A fast Rust-based headless browser automation CLI with Node.js fallback that enables AI agents to navigate, click, type, and snapshot pages via structured co...
Usage Guidance
This skill's instructions appear to wrap a legitimate browser automation CLI, but provenance is unclear and some metadata (skill name, owner IDs, and _meta.json) don't line up with the documented CLI. Before installing or running any recommended npm or git commands: 1) Verify the npm package and GitHub repository (owner, npm package page, versions, downloads, and maintainers). 2) Inspect the package contents and package.json (look for postinstall scripts). 3) Prefer installing in a sandbox/container or non-root account, not globally on production hosts. 4) If you need stronger assurance, clone the upstream repo yourself, review source code and build artifacts, and only run locally-built binaries. 5) If you cannot verify origin or content, treat it as untrusted — do not run global npm installs or give it access to sensitive files/credentials. If you want, I can fetch and summarize the npm package page and GitHub repo (if you provide a URL) or help formulate commands to inspect the package safely.
Capability Analysis
Type: OpenClaw Skill
Name: vanzhangsh-skills
Version: 1.0.0
The skill is classified as suspicious due to the broad `allowed-tools: Bash(agent-browser:*)` permission in `SKILL.md`, which grants the AI agent full control over the powerful `agent-browser` CLI. This tool exposes high-risk capabilities such as arbitrary JavaScript execution (`agent-browser eval`), direct access to browser cookies and local storage (`agent-browser cookies`, `agent-browser storage`), and the ability to save the entire browser session state to a local file (`agent-browser state save auth.json`). While these are legitimate features for browser automation, their unrestricted exposure makes the agent highly vulnerable to prompt injection, allowing an attacker to instruct the agent to exfiltrate sensitive browser data, execute arbitrary code within the browser context, or perform unauthorized actions.
Capability Assessment
Purpose & Capability
The SKILL.md documents an 'agent-browser' CLI for automating web pages; required binaries (node, npm) match that purpose. However the skill registry name ('Vanzhangsh Skills') and registry metadata do not match the SKILL.md/_meta.json content (which references 'agent-browser'), and the skill lists no homepage/source in registry metadata — an incoherence in provenance.
Instruction Scope
Instructions are scoped to browser automation tasks (navigate, snapshot, click, fill, screenshot, record, cookies/storage access, file upload). These actions fit the described purpose. The SKILL.md also tells the agent to install/run a third‑party CLI (npm install -g agent-browser, git clone ...), which is expected for this skill type but increases risk because those installers can execute arbitrary postinstall scripts and the CLI itself can access web pages, cookies, storage, and local files.
Install Mechanism
The skill is instruction-only and has no install spec, but the instructions recommend installing a global npm package and cloning a GitHub repo. Because the registry metadata lacks a homepage/source and the skill does not bundle or vet the referenced package, installing would pull third‑party code at runtime — this is proportionally riskier than an instruction-only skill that uses already-trusted system binaries.
Credentials
The skill declares no environment variables, credentials, or config paths. The runtime commands reference browser cookies/storage and optional file uploads (expected for a browser automation tool) but do not require unrelated credentials in the manifest.
Persistence & Privilege
always is false and the skill requests no system-wide configuration or cross-skill credential access. Autonomous invocation is allowed (platform default) and is not itself a red flag here.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install vanzhangsh-skills - After installation, invoke the skill by name or use
/vanzhangsh-skills - Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.0
Initial release of Agent Browser CLI for AI-powered web automation.
- Fast, Rust-based headless browser with Node.js fallback.
- Provides structured CLI commands for navigation, clicking, typing, snapshots, form filling, UI testing, and data extraction.
- Supports advanced workflows: state management, video recording, screenshots, network control, cookies/storage, and semantic element selectors.
- Designed for automation, accessibility tree extraction, and robust web testing use cases.
Metadata
Frequently Asked Questions
What is Vanzhangsh Skills?
A fast Rust-based headless browser automation CLI with Node.js fallback that enables AI agents to navigate, click, type, and snapshot pages via structured co... It is an AI Agent Skill for Claude Code / OpenClaw, with 402 downloads so far.
How do I install Vanzhangsh Skills?
Run "/install vanzhangsh-skills" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Vanzhangsh Skills free?
Yes, Vanzhangsh Skills is completely free (open-source). You can download, install and use it at no cost.
Which platforms does Vanzhangsh Skills support?
Vanzhangsh Skills is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Vanzhangsh Skills?
It is built and maintained by vanzhangsh (@vanzhangsh); the current version is v1.0.0.
More Skills