Bowser is an agentic browser automation and UI testing system built on a composable, four-layer architecture (Skill, Subagent, Command, Justfile). It addresses the need for consistent, dual-purpose tooling, offering both observable (real Chrome) and headless (isolated Chromium) modes. This system benefits engineers and power users by providing a flexible, scalable, and independently testable framework for repeatable browser automation.

How It Works

The system comprises four layers: Skills provide browser capabilities (Playwright CLI or Chrome MCP), Subagents scale these into parallel, isolated sessions, Commands orchestrate workers into repeatable workflows, and the Justfile offers single-command execution. This layered design emphasizes composability and independent testability, allowing users to engage at any level—from testing individual skills to running full orchestrations. The core advantage is its adaptability for diverse browser automation and UI testing use cases.

Quick Start & Requirements

Install Claude Code via curl -fsSL https://claude.ai/install.sh | bash or Homebrew (brew install --cask claude-code). Navigate to the project (cd bowser) and start Claude Code (claude Play). The Playwright skill requires playwright-cli (npm install -g playwright-cli@latest). The optional just tool (brew install just) simplifies execution via just commands.

Highlighted Details

• Four-Layer Architecture: Skill, Subagent, Command, and Justfile enable composability, orchestration, scaling, and reusability.
• Dual-Purpose Skills: Claude-Bowser (observable, personal workflows, real Chrome) and Playwright-Bowser (headless, isolated Chromium for scalable UI testing).
• Independent Testability: Each layer is designed for direct testing, facilitating robust development.
• YAML-Defined Stories: User workflows and test cases are defined in YAML files, automatically discovered.
• Token Efficiency: Playwright-Bowser uses efficient CLI interaction; Claude-Bowser uses more verbose MCP tool schemas.

Maintenance & Community

Associated with IndyDevDan, with learning resources on a YouTube channel for agentic coding. No specific community channels or contributor details are provided.

Licensing & Compatibility

The provided README content does not specify a software license, making commercial use compatibility undetermined.

Limitations & Caveats

Claude-Bowser is single-instance, not CI-friendly, requires --chrome, and uses the user's real browser. Playwright-Bowser is Chromium-only, lacks access to real browser profiles/extensions, and requires explicit authentication handling.