Clawhub Publish 146198

Automate web browsing tasks like navigation, form filling, clicking, and data extraction using a fast Rust-based headless browser with Node.js fallback.

Security Scan

The skill's purpose (headless browser CLI) is plausible, but inconsistent metadata and the instruction to globally install an unreviewed npm package create risk and don't fully match the bundle's public data.

Purpose & Capability

The described capability (browser automation via an 'agent-browser' CLI) is coherent with the skill's description and examples. However, the registry metadata/packaging labels in the bundle do not match the SKILL.md (registry slug is 'clawhub-publish-146198' while the SKILL.md and _meta.json refer to 'agent-browser' and the ownerId values differ), which raises provenance and supply-chain questions.

Instruction Scope

SKILL.md only instructs installing and invoking an external CLI (agent-browser) and gives commands for navigation, fill, click, snapshot. The instructions do not ask the agent to read unrelated files or env vars. Still, the runtime behavior of the CLI will be able to interact with pages that include credentials and sensitive data, so the instructions implicitly enable capture/exfiltration if the CLI is malicious or misused.

Install Mechanism

There is no formal install spec in the bundle; the SKILL.md asks the user to run 'npm install -g agent-browser'. Installing an arbitrary npm package globally is a moderate-risk operation because npm packages may run lifecycle scripts or download binaries. The skill does not provide a homepage, source repo, checksum, or verified release URL to validate the package before installing.

Credentials

The skill declares no required environment variables or credentials, which matches the SKILL.md. However, the tool's intended use (automating logins/forms) means it will see any credentials the user enters into pages it automates — the skill does not explicitly declare how such data is handled or transmitted.

Persistence & Privilege

The skill does not request always:true and does not claim special persistent privileges. It's user-invocable and may be used autonomously by the agent (default), which is normal for skills; no extra persistent system-wide changes are indicated in the bundle.

What to consider before installing

This skill instructs you to globally install an npm package (agent-browser) but provides no homepage or source to inspect. Before installing: 1) verify the package on the npm registry and check the package owner, repository link, and recent publish history; 2) review the package source code or repository for postinstall scripts or downloads; 3) avoid global install as root — prefer a sandbox/container or scoped environment; 4) be cautious about automating logins or sensitive pages with untrusted CLI tools (they can capture credentials); and 5) ask the publisher for consistent metadata (matching slug/owner) and a canonical homepage/source. The metadata mismatch (different slugs/ownerIds between registry and _meta.json/SKILL.md) is the main red flag here.

Agent Browser

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 commands.

When to use

• Automating web interactions (login, form fill, scraping)
• Extracting structured data from pages
• Testing web UIs or workflows
• Filling forms programmatically

Prerequisites

• Node.js ≥ 18
• npm install -g agent-browser (run once to install CLI)

How it works

1. agent-browser open <url> → launch browser & navigate
2. agent-browser snapshot -i → get interactive elements with stable refs (e.g., @e1)
3. Use refs to interact: click @e1, fill @e2 "text", wait @e3
4. agent-browser close → clean up

Key capabilities

• Snapshot with accessibility tree or interactive-only mode
• Precise element interaction (click/fill/hover/drag/upload)
• Wait conditions (element visible, URL change, network idle)
• Screenshot, PDF export, video recording
• Cookie/storage/network control
• Tab/window/frame management
• JSON output for programmatic parsing

Example

agent-browser open https://example.com/login
agent-browser snapshot -i
agent-browser fill @e1 "[e-mail]"
agent-browser fill @e2 "password123"
agent-browser click @e3
agent-browser wait --url "/dashboard"