gstack

Provides structured AI agent workflow skills and a persistent headless Chromium browser for software development.

2|Updated Mar 26, 2026
One-click install
npx skills add https://github.com/westkite1201/garden-engine --skill gstack-westkite1201
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: gstack
Source: https://github.com/westkite1201/garden-engine/tree/main/.claude/skills/gstack
Command: npx skills add https://github.com/westkite1201/garden-engine --skill gstack-westkite1201

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires playwright, bun, and includes scripts (resource) components.

What problem does it solve? AI coding agents lack structured engineering workflows and fast, stateful browser automation, leading to unreviewed code, skipped QA, and slow multi-second browser commands that lose login state between calls. ## Core Features & Use Cases - Specialist workflow skills: Over 20 slash-command skills covering CEO-level plan review, engineering review, design audits, systematic debugging, QA in a real browser, PR shipping, security audits, and retrospectives. - Persistent headless browser: A compiled CLI talks to a long-lived Chromium daemon over localhost HTTP, giving ~100ms commands, persistent cookies and tabs, ref-based element selection via the accessibility tree, and cookie import from real browsers. - Safety guardrails: Skills like /careful, /freeze, and /guard warn or block destructive commands and restrict edits to specific directories. - Use Case: A developer runs /plan-eng-review to lock architecture, /qa to open a real browser and verify the app, then /ship to run tests, review, and open a PR — all as guided AI workflows. ## Quick Start Ask the agent to run the /office-hours skill to reframe your product idea before writing any code.

Frequently Asked Questions about gstack

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I add structured code review workflows to Claude Code?

Install gstack and invoke skills like /review for pre-landing PR review or /plan-eng-review for architecture review. Each skill is a SKILL.md file that gives the agent a specialist role with defined steps and quality gates.

How do I give an AI agent a fast headless browser?

Use the gstack browse CLI, which talks to a persistent Chromium daemon over localhost HTTP. The first call takes about 3 seconds to start the server; subsequent commands complete in roughly 100-200ms with cookies and tabs preserved.

What is the difference between gstack browse and Playwright MCP?

gstack browse is a compiled CLI that outputs plain text with zero protocol overhead per call, while MCP tools add JSON schema framing costing thousands of tokens per command. The daemon model also keeps browser state alive between commands.

Does gstack browser automation work on Windows and Linux?

Yes, the browse daemon runs on macOS, Linux, and Windows after platform-specific fixes for process detachment and Chromium sandboxing. Cookie decryption from real browsers supports macOS Keychain and Linux GNOME Keyring.

Why do element refs become stale after page changes?

Refs are cleared on navigation and can go stale when SPAs mutate the DOM without navigating. Run snapshot again to get fresh refs; the resolveRef check fails fast with a clear error instead of waiting for a timeout.

How do I prevent an AI agent from running destructive commands?

Activate the /careful skill to warn before commands like rm -rf or force-push, /freeze to lock edits to one directory, or /guard to enable both at once. These are advisory guardrails built into the skill set.