Skip to content

Engineering Playbook

Ship Products 810× Faster With AI-Guided Engineering

0.0(0 reviews)by at garrytan

Published May 2026

Complete AI engineering system: 23 specialist skills + 8 tools in a proven sprint workflow. Think, plan, build, review, test, ship, reflect.

Product Ideation & PlanningDesign System & Visual DevelopmentCode Review & QualityQA & TestingShipping & DeploymentDeveloper IntelligenceSafety & UtilitiesKnowledge Integration

Install Command

Run this command to deploy the playbook to your environment.

Included Skills

Reframe Ideas & Lock In Plans Before Code

YC-style product interrogation and multi-layer planning. Five skills that challenge assumptions, rethink scope, and build consensus before you write a line of code.

/office-hours/plan-ceo-review/plan-design-review/plan-eng-review/plan-devex-review

Build Design Systems & Production HTML from Scratch

Four complementary design skills: system design, visual exploration, code generation, and live-site audits. From concept to shipped HTML in one workflow.

/design-consultation/design-shotgun/design-html/design-review

Catch Production Bugs & Security Issues Before Shipping

Structural code review with bug-finding, security audits, and independent second opinions from multiple AI models.

/review/investigate/cso/codex

Deploy & Verify From Branch to Production in One Command

Release engineering automation: test bootstrap, coverage audits, PR creation, merging, deploying, and production health verification.

/ship/land-and-deploy/document-release/setup-deploy

Learn, Remember, Measure, & Improve Across Sessions

Persistent memory for learned patterns, session state restoration, team retros, and developer profile tracking.

/learn/retro/context-save/context-restore

Add Guardrails, Unlock Real Browsers & Keep gstack Updated

Four safety skills, a real browser launcher, and self-upgrade capability.

/careful/freeze/guard/open-gstack-browser

Sync Codebase Knowledge Across Machines With GBrain

Optional persistent knowledge base for semantic code search and cross-machine memory sync.

/setup-gbrain/sync-gbrain

The Problem

You're shipping more products than ever, but the process is chaotic. Without structured workflows, AI agents hallucinate, introduce bugs, and skip testing. Without guidance, they treat your code like a blank canvas instead of respecting existing patterns. You need a system that turns Claude Code into a virtual engineering team.

Step-by-Step Workflow

  1. 1
    Claude logo
    Claude

    Step 1 using Claude

  2. 2
    P
    Playwright

    Step 2 using Playwright

  3. 3
    B
    Bun

    Step 3 using Bun

  4. 4
    G
    Git

    Step 4 using Git

  5. 5
    C
    Chromium

    Step 5 using Chromium

  6. 6
    S
    Supabase

    Step 6 using Supabase

  7. 7
    Grain logo
    Grain

    Step 7 using Grain

  8. 8
    O
    OpenAI Codex

    Step 8 using OpenAI Codex

  9. 9
    Conductor logo
    Conductor

    Step 9 using Conductor

How It Works

After you install gstack, every Claude Code session loads 23 specialist skills organized as a sprint: `/office-hours` reframes your product idea, `/plan-*` skills lock in architecture and design, `/review` finds production bugs before they ship, `/qa` tests with a real browser, `/ship` deploys with confidence. Each skill feeds into the next. Nothing falls through the cracks.

The Biggest Win

Run 10+ parallel product sprints at once, shipping complete features with architecture, design, testing, and QA—all from one workflow.

Watch Out For

Requires Claude Code (not just Claude API). All 23 skills run as slash commands in Claude Code context. gstack is opinionated—it enforces testing, security reviews, and structured planning. If you prefer ad-hoc prompting, this will feel prescriptive. Parallel sprints scale to 10-15 at a time with Conductor; beyond that you manage context switching manually.

Under the Hood

Written in TypeScript/Bash. Browse daemon uses Playwright + Chromium (auto-detects Bun or falls back to Node.js; Windows requires both). GBrain integration is optional (local PGLite or remote Supabase). Telemetry is opt-in, zero code/paths sent. Supports Claude, Codex, OpenCode, Cursor, Factory Droid, Slate, Kiro, Hermes, GBrain via a single `./setup` script.

Tools in this Playbook

Claude logo
Claude
Playwright
Bun
Git
Chromium
Supabase
Grain logo
Grain
OpenAI Codex
Conductor logo
Conductor

Key Features

Sprint Workflow

23 specialist skills organized as a complete sprint: product ideation, planning (CEO/eng/design/DX), code review, QA, deployment, and retrospectives. Each skill feeds into the next with zero manual handoff.

Real Browser Testing

Persistent Chromium daemon (`/browse`) with 100+ commands: navigate, fill forms, take screenshots, check console, verify state. ~100ms per command. Browser state persists between calls.

Production-Ready QA

`/qa` systematically tests your app: reads git diff, identifies affected pages, finds and fixes bugs with atomic commits, auto-generates regression tests.

Multi-Layer Review

`/review` finds bugs that pass CI but fail in prod. `/codex` brings independent perspective from OpenAI. `/cso` runs OWASP + STRIDE security audits. Cross-model findings for high confidence.

Design System & HTML

`/design-consultation` builds design systems from scratch. `/design-shotgun` generates 3+ visual variants with taste memory. `/design-html` renders production HTML with Pretext for responsive layouts.

About This Playbook

License
MIT
Industry
Technology
Skills
2 workflows, 6 sub-skills, 23 standalone