GStack

Product Introduction

1. Definition: GStack is an advanced workflow orchestration framework designed specifically for Claude Code, the command-line interface (CLI) agent from Anthropic. It functions as a collection of "opinionated workflow skills" implemented through a combination of Markdown-based prompt engineering and native binaries. Technically, it is a local-first enhancement suite that categorizes AI interactions into specific cognitive modes, utilizing the Bun runtime and Playwright-driven browser automation to extend the functional capabilities of the Claude Code environment.

2. Core Value Proposition: GStack exists to solve the "generic assistant" problem where AI agents provide inconsistent, mid-level responses by default. By providing "explicit gears" for different stages of the Software Development Lifecycle (SDLC), GStack allows developers to summon specialized personas—Founder, Engineering Manager, Staff Engineer, and QA Engineer—on demand. Its primary value lies in increasing shipping velocity and code quality through high-rigor, role-specific slash commands that integrate directly into the developer's terminal workflow.

Main Features

1. /browse (QA Engineer Mode): This is a high-performance browser automation skill that gives Claude Code "eyes." Unlike standard MCP tools, GStack uses a compiled native binary built on Microsoft Playwright that interfaces with a persistent Chromium daemon. This allows the agent to navigate web applications, handle authentication/cookies, take screenshots, and inspect the DOM with sub-second latency (100-200ms per call after initial startup). It enables automated end-to-end testing, visual regression checks, and documentation reading directly from the CLI.

2. /plan-ceo-review (Founder Mode): This skill shifts the AI's focus from literal code implementation to product strategy and user empathy. It is designed to "pressure-test" requests by asking if the user is building the right product. It prioritizes "10-star" product thinking—identifying features that offer high leverage, such as auto-identifying products from images or suggesting pricing comps—rather than simply adding buttons or forms.

3. /plan-eng-review (Tech Lead Mode): Once product direction is set, this command focuses on technical architecture and system boundaries. It forces the model to generate structured technical specifications, including sequence diagrams, state machines, data-flow diagrams, and test matrices. By requiring visual representations of the system, it uncovers hidden technical assumptions and architectural edge cases that text-only planning often misses.

4. /review (Staff Engineer Mode): This is a "paranoid" code auditing tool that looks beyond basic syntax and style. It specifically hunts for production-breaking structural issues such as N+1 queries, race conditions in concurrent environments, trust boundary violations, and stale read patterns. It acts as a final gatekeeper that simulates a high-seniority human peer review focused on system stability and security.

5. /ship (Release Engineer Mode): This command automates the "last mile" of development. It handles repetitive release hygiene, including syncing the local branch with the main repository, running test suites, checking for inconsistent branch states, updating version metadata or changelogs, and pushing to the remote to open a Pull Request. It is optimized for maintaining momentum by removing the cognitive load of manual Git operations.

6. /retro (Engineering Manager Mode): An analytical tool that parses commit history and local metadata to generate data-driven engineering retrospectives. It computes metrics such as lines of code (LOC) per day, test-to-code ratios, PR sizes, and shipping streaks. It saves JSON snapshots in a .context/retros/ directory, allowing for longitudinal trend tracking and performance comparisons across different weeks.

Problems Solved

1. Pain Point: AI Inconsistency and "Mushy" Responses. Standard AI agents often provide a generic blend of advice that is neither strategically ambitious nor technically rigorous. GStack enforces "cognitive modes," ensuring that planning is distinct from reviewing and that review is distinct from execution.

2. Target Audience: Senior Software Engineers, Startup Founders, High-Velocity Product Teams, and power users of the Claude Code CLI. It is specifically marketed as "not a prompt pack for beginners," but rather an operating system for experienced developers who ship production-grade software daily.

3. Use Cases:

• Rapid Prototyping: Using /plan-ceo-review to refine an MVP idea before a line of code is written.
• Automated QA: Running /browse on a staging environment to verify a signup flow before merging a PR.
• Technical Debt Reduction: Using /review to find architectural flaws in legacy codebases that pass standard CI/CD lints.
• Team Synchronization: Adding GStack to a project's .claude/skills folder so all teammates have access to identical, high-rigor review and shipping workflows.

Unique Advantages

1. Differentiation: Unlike generic AI coding assistants or simple prompt libraries, GStack is a toolchain that includes compiled binaries for browser interaction. It moves the AI from a "text-in, text-out" paradigm to an "action-oriented" paradigm where the agent can actually see the UI and interact with the live application it just built.

2. Key Innovation: The persistent browser session. By keeping a Chromium instance running in the background, GStack eliminates the overhead of launching a browser for every individual command. This makes "Agentic QA" viable in a real-time terminal loop, allowing for 60-second full-site audits that would take a human 10-15 minutes to perform manually.

Frequently Asked Questions (FAQ)

1. How do I install GStack for Claude Code? To install GStack, you must have Claude Code, Git, and Bun v1.0+ installed. Run the git clone command into your ~/.claude/skills/gstack directory and execute the ./setup script. This will compile the browser binary and create the necessary symlinks so Claude can recognize the new slash commands.

2. What is the difference between /plan-ceo-review and /plan-eng-review? /plan-ceo-review focuses on "Founder Mode" or product taste—deciding what should be built based on user value and strategy. /plan-eng-review focuses on "Tech Lead" rigor—determining how to build it using diagrams, architecture specs, and edge-case analysis.

3. Is GStack safe to use with production environments? The /browse tool runs a real Chromium session that can handle cookies and session states. While highly effective for QA, users should exercise caution when using it against sensitive production environments, as the agent has the capability to perform real actions like clicking buttons or submitting forms just like a human user.

4. Does GStack work on Windows? Currently, GStack is optimized for macOS and Linux (supporting both x64 and arm64 architectures). It requires a Unix-like environment to run the shell scripts and manage the native browser binaries effectively.

5. How does the /retro command track my progress? /retro analyzes your local Git commit history and work patterns. It saves these insights as JSON snapshots in the .context/retros/ folder within your project. This allows you to run comparative analyses (e.g., /retro compare) to see how your shipping velocity or test coverage has changed over time.