core
Core provides a high-performance command-line interface built in Rust that enables AI agents to control web browsers programmatically. Automate navigation, element interaction, form submission, and data extraction with a streamlined CLI designed for agent workflows. Install globally or as a project dependency to add browser automation capabilities to your AI systems.
Core is a high-performance command-line interface built in Rust that enables AI agents to control web browsers programmatically. Core automates website interactions by providing streamlined CLI commands for navigation, element clicking, form filling, and data extraction. Instead of raw HTML, core uses compact accessibility-tree references to read page content and take screenshots, making it ideal for AI agent workflows that need efficient, structured browser control.
AI-generated summary based on this skill's SKILL.md
Decision gist · record as of 2026-07-28
Core is a high-performance command-line interface built in Rust that enables AI agents to control web browsers programmatically. Core automates website interactions by providing streamlined CLI commands for navigation, element clicking, form filling, and data extraction. Instead of raw HTML, core uses compact accessibility-tree references to read page content and take screenshots, making it ideal for AI agent workflows that need efficient, structured browser control.
Use it when
- Core provides CLI commands that let you interact with web page elements using accessibility-tree references rather than fragile selectors.
- Yes, core can take screenshots of webpages programmatically via its CLI.
Verify before relying
Read SKILL.md below before installing (13 files). Open directory: indexed for reading, not audited.
Install
vercel-labs/agent-browser/core · repository language: Rust
Open directory. Skills are indexed for reading, not audited. Review a skill's body before installing it.
Frequently asked questions
AI-generated answers based on this skill's SKILL.md and metadata
What is core and how does it automate website interactions?
Core is a high-performance command-line interface built in Rust that enables AI agents to control web browsers programmatically. Core automates website interactions by providing streamlined CLI commands for navigation, element clicking, form filling, and data extraction. Instead of raw HTML, core uses compact accessibility-tree references to read page content and take screenshots, making it ideal for AI agent workflows that need efficient, structured browser control.
How do I fill a form and click a button with core?
Core provides CLI commands that let you interact with web page elements using accessibility-tree references rather than fragile selectors. To fill a form and click a button, you reference elements by their compact accessibility identifiers, then issue commands to populate form fields and trigger clicks. This approach keeps your automation robust across page layout changes and integrates seamlessly into agent scripts and MCP server environments.
Can core take a screenshot of a webpage programmatically?
Yes, core can take screenshots of webpages programmatically via its CLI. When you capture a screenshot with core, it also provides an accessibility-tree snapshot that maps interactive elements to compact references. This dual output—visual snapshot plus structured element references—lets AI agents both see the page and interact with it using reliable, machine-readable identifiers instead of parsing raw HTML.
How does core handle authentication and persist browser session state across multiple commands?
Core maintains browser session state across multiple CLI commands, allowing you to authenticate once and then perform subsequent interactions without re-logging in. This persistent session handling is essential for multi-step workflows where an AI agent needs to log in, navigate authenticated pages, and extract data across several command invocations. Core's session persistence simplifies complex automation scenarios that would otherwise require managing cookies and tokens manually.
What is the core agent-browser snapshot and refs tutorial approach?
Core's snapshot and refs approach captures both a visual screenshot and an accessibility-tree representation of the page, with each interactive element assigned a compact reference identifier. This tutorial-friendly pattern lets AI agents understand page structure without parsing HTML directly. References remain stable across commands, enabling reliable element targeting and reducing failures from stale selectors or dynamic content changes.
How do I integrate core into an MCP server or AI agent sandbox?
Core can be installed globally or as a project dependency to add browser automation capabilities to your AI systems. For MCP server integration, core's CLI commands map naturally to structured tool definitions, allowing agents to call browser actions through standardized interfaces. This integration pattern works seamlessly in eve agent sandboxes and other structured tool environments, giving AI systems reliable, permission-bounded access to web automation without embedding a full browser library.
SKILL.md
Rendered from the published skill. Quoted content, verbatim.
agent-browser core
Fast browser automation CLI for AI agents. Chrome/Chromium via CDP, no Playwright or Puppeteer dependency. Accessibility-tree snapshots with compact @eN refs let agents interact with pages in ~200-400 tokens instead of parsing raw HTML.
Most normal web tasks (navigate, read, click, fill, extract, screenshot) are covered here. Load a specialized skill when the task falls outside browser web pages — see When to load another skill.
The core loop
```bash agent-browser open
(truncated - see the full file via the links below)
File tree — 13 files
skill-data/core/SKILL.md
skill-data/core/references/authentication.md
skill-data/core/references/commands.md
skill-data/core/references/profiling.md
skill-data/core/references/proxy-support.md
skill-data/core/references/session-management.md
skill-data/core/references/snapshot-refs.md
skill-data/core/references/trust-boundaries.md
skill-data/core/references/video-recording.md
skill-data/core/references/webgpu.md
skill-data/core/templates/authenticated-session.sh
skill-data/core/templates/capture-workflow.sh
skill-data/core/templates/form-automation.sh
Let your AI agent find skills like this
Example. Real query, live index.
You found this page by searching. An agent finds it by wishing: SkillFed indexes 56,283 agent skills by what they can do, searchable in plain language.
wish › “Automate web browser interactions (navigate, click, fill forms, extract data) via command-line interface for AI agents”
Give your agent the search over MCP, or paste the wish link into any chat. No install? Search from any chat →
Related skills
agent-browser enables programmatic control of web browsers, letting you automate clicks, text input, page navigation, and other user interactions at scale. Built for integration with AI agents and automation frameworks, it simplifies complex browser workflows without manual intervention.
This skill equips your AI agents with the ability to control web browsers programmatically, handling clicks, form submissions, and page navigation as part of automated workflows. Built on a Go-based engine that communicates via WebSocket, it integrates seamlessly into multi-turn agent conversations and supports custom tool extensions. Perfect for automating repetitive web tasks or building intelligent browser automation into your agent applications.
Agent Browser equips your AI agent with the ability to explore and interact with web content in real time. Navigate between pages, retrieve information, and perform actions on websites as if a human were at the controls. Perfect for agents that need autonomous web research and data gathering.
Agent Browser gives AI systems the ability to independently navigate websites, extract information, and interact with web elements without human intervention. This skill bridges the gap between AI reasoning and real-world web data, enabling agents to perform research, monitoring, and automated workflows across the internet.
Agent Browser empowers your AI agents to independently navigate and interact with web content, handling everything from page navigation to data extraction and form submission. Perfect for automating research, monitoring, and web-based task execution without manual intervention.
Agent Browser equips your AI agents with full web navigation and interaction abilities. Browse websites, click links, fill forms, and extract live content—all programmatically integrated into your agent's workflow. Perfect for research, data gathering, and dynamic web-based tasks.
More skills agent-browser (Apache-2.0) · agent-browser (MIT) · Agent Browser (NOASSERTION) · electron (Apache-2.0) · dev-browser (MIT)