hubcap

Hubcap — Claude Code Skill

A Claude Code skill for Hubcap, a Go CLI that wraps the entire Chrome DevTools Protocol in 118 composable shell commands.

Why Hubcap?

Full CDP coverage. 118 commands spanning navigation, DOM queries, form filling, screenshots, network interception, device emulation, accessibility audits, performance profiling, and more. If you can do it in Chrome DevTools, you can script it with hubcap.

Designed for agents. Every decision prioritises AI agent usability: JSON output by default, semantic exit codes (0 success, 1 error, 2 connection failed, 3 timeout), a self-describing interface (hubcap help <command>), and stateless architecture that eliminates session management. Commands compose naturally with Unix pipes and shell chaining, keeping token consumption low.

Context friendly. The skill keeps a small footprint — just command categories and key patterns. When the agent needs detailed usage for a specific command, it runs hubcap help <command> to pull full documentation on demand. No context burned on 118 command docs upfront.

Single binary, no dependencies. Just brew install and go. No runtimes, no SDKs, no page objects to construct.

Prerequisites

Install hubcap:

brew install tomyan/tap/hubcap

Or download a binary from the latest release.

Install the skill

Add the marketplace and install:

/plugin marketplace add tomyan/claude-skills
/plugin install hubcap@tomyan-skills

Or install directly:

git clone https://github.com/tomyan/claude-skill-hubcap.git ~/.claude/plugins/hubcap

What the agent can do

• Navigate to pages and wait for them to load
• Fill out forms and click buttons
• Take screenshots and export PDFs
• Scrape text, links, tables, and structured data
• Extract content without vision (HTML, accessibility trees, DOM snapshots)
• Run accessibility audits
• Monitor network traffic and intercept requests
• Emulate mobile devices and set viewport sizes
• Execute JavaScript and evaluate expressions
• Assert page state for testing
• Capture performance metrics and heap snapshots

Example

Ask the agent something like:

Take a screenshot of https://example.com on an iPhone 12

The agent will run:

hubcap setup launch
hubcap emulate iPhone-12
hubcap goto --wait https://example.com
hubcap screenshot --output example-iphone12.png

Links

• Introducing Hubcap — blog post on the motivation and design
• Hubcap documentation
• Hubcap skill page
• Hubcap source