VividKit Crystal clear AI coding
VI
Choose Your Setup Get Started with AgentKit Continue with ClaudeKit Migrate from ClaudeKit to AgentKit Clean cutover from ClaudeKit What Is AgentKit? How AgentKit Works AgentKit Kits, Skills, Agents & Hooks AgentKit Runtime Support AgentKit Desktop App AgentKit Configuration AgentKit CLI Quick Reference AgentKit Skills Cheatsheet AgentKit Workflows AgentKit Statusline Update AgentKit CLI & Kits AgentKit Helper Clean reinstall of AgentKit AgentKit Troubleshooting What is ClaudeKit? ClaudeKit CLI Guide Command Finder Commands Cheat Sheet Workflow Recipes CLI Quick Reference Claude Mechanics Permission Modes ClaudeKit Hooks Coexistence ClaudeKit x Codex CK → Other coding tools How ClaudeKit Works Inside ClaudeKit Finding Your Unknowns Session Recovery Fix from Logs UI/UX Guide Accounts Switcher IDE Config Remote Control Codex App System Diagramming UI Review Gate Promotions Support VividKit

Quick Ref / Command

Engineer Kit

/ak:agent-browser

Profile-independent browser automation

Uses the agent-browser CLI for autonomous browsing, snapshots, screenshots, form fills, scraping, Electron apps, Slack workflows, bug hunts, cloud browsers, and QA when real Chrome profile state is not required.

01

Choose the browser owner

02

Install and verify

03

Load live skills

04

Pick specialty

Rule 01

Agent-browser is the workflow source: load skills get core instead of copying stale command details.

Rule 02

Prefer this skill for profile-independent browser work; prefer ak:chrome-profile for real cookies and logged-in state.

Rule 03

Compact accessibility snapshots and stable refs are the main reliability lever.

Rule 04

Cloud, Electron, Slack, and exploratory QA each have specialized live workflow content.

Execution Map

Core lanes from input to output, aligned with the command's real execution path.

01

Start

Input, scope, route

  1. 1 Choose the browser owner Use agent-browser for fresh or tool-managed sessions; switch to ak:chrome-profile when cookies, accounts, tenants, or existing real Chrome state matter.
  2. 2 Install and verify Install or upgrade with npm, download Chromium once, and confirm the binary version before relying on commands.
02

Work

Agent / skill execution

  1. 3 Load live skills Run agent-browser skills get core before command work so snapshots, refs, troubleshooting, and command details match the installed binary.
  2. 4 Pick specialty Load electron, slack, dogfood, vercel-sandbox, or agentcore workflow content when the task is not ordinary web-page automation.
03

Verify

Gate, review, validation

  1. 5 Run with compact refs Navigate, snapshot, interact through accessibility-tree element refs, and re-snapshot after page changes to avoid stale targets.
  2. 6 Observe and debug Use screenshots, video/session state, the dashboard on port 4848, and documented troubleshooting for missing Chromium, stale sessions, or missing elements.
04

Close

Report, handoff, artifact

  1. 7 Report limits Name when a task needs the user's real logged-in Chrome or low-level CDP diagnostics instead of pretending the managed browser can see that state.

Syntax, arguments, subcommands

How to invoke the skill: syntax, positional arguments, shared options, then each subcommand with its own syntax and outcome.

Syntax

/ak:agent-browser [url or task]

Arguments

[url or task] Required

URL or browser task

A URL to inspect, or a natural-language task with the target, expected result, allowed actions, artifacts to keep, and cleanup rule.

/ak:agent-browser "Open https://staging.example.com, inspect the checkout flow, capture screenshots, do not submit an order, and close the session"

Shared options

-p browserbase

Browserbase provider

Use a remote Browserbase browser after credentials, access, and provider cost are approved. Do not fall back silently if it is unavailable.

/ak:agent-browser "Use Browserbase (-p browserbase) to open https://example.com, capture the landing page state, then close the remote session"

Special Operations

Live core reference

agent-browser skills get core is the starting point for workflows and troubleshooting.

Profile split

Real Chrome cookies and account state belong to ak:chrome-profile, not this skill.

Dashboard

Port 4848 exposes session observability without exposing per-session browser ports.

Sample Prompt

Concrete invocations for each flag, subcommand, mode, or route available in the live workflow.

Bounded browser QA Recommended 
/ak:agent-browser "Open https://staging.example.com, inspect the checkout flow through the confirmation preview, capture screenshots at each step, do not submit an order, and close the session"
Use when:
Use when a fresh or tool-managed browser is enough for a long interaction, screenshots, and exploratory QA.
Expected:
Confirms permitted actions, loads agent-browser skills get core, drives the flow with snapshots and @eN refs, reports screenshots, final page state, skipped submission, and session closure.
Inspect one URL 
/ak:agent-browser https://example.com
Use when:
Use when the task starts with opening and inspecting a public page in a profile-independent session.
Expected:
Opens the URL in the CLI-managed Chromium session, takes compact accessibility snapshots, refreshes refs after changes, and reports observations without using the user’s real Chrome cookies.
Cloud browser run 
/ak:agent-browser "Use Browserbase (-p browserbase) to open https://example.com, capture the landing page state, then close the remote session"
Use when:
Use for CI or environments without a local browser after Browserbase credentials and cost are approved.
Expected:
Uses the documented Browserbase provider path, records the cloud provider and target origin, captures requested evidence, and closes the remote browser so provider resources are not left running.
Specialized workflow 
/ak:agent-browser "Load the Slack agent-browser workflow, verify the notification flow in the approved workspace, capture evidence, and avoid sending messages"
Use when:
Use when Slack, Electron, dogfood QA, Vercel Sandbox, or AgentCore requires the installed specialized workflow content.
Expected:
Loads the relevant agent-browser skills get workflow before acting, keeps interactions inside the approved app or workspace, captures reviewable evidence, and reports any limits or skipped mutations.

Handled Scope

  • Autonomous browsing
  • Form filling
  • Screenshots and QA
  • Electron and Slack automation
  • Browserbase/cloud sessions

Next