Interrogate Ideas Before Building8 skills
Force critical thinking on product assumptions before writing code
Engineering Blueprint
Run Software Development End to End
garrytanMay 2026
A 40+ AI skill workflow for product ideation, planning, design, review, QA, deployment, and retrospectives.
Interrogate IdeasPlan with LeadershipPlan Technical DetailsDesign & IterateReview CodeTest & AuditShip & DeployReflect & LearnUtilities & Integration
Install Command
Run this command to deploy the blueprint to your environment.
What problem does this solve?
Runs your complete product development cycle as one connected workflow instead of disconnected tasks. Each skill feeds the next with context and artifacts: design docs from /office-hours flow into /plan-ceo-review, which informs /plan-eng-review, which feeds /qa with test plans, which feeds /ship with verified changes. Because skills share context, nothing falls through cracks and every step builds on what came before.
How does it work?
Load gstack in Claude Code. (1) Start with /office-hours or /spec to interrogate your idea. (2) Run /plan-ceo-review, /plan-eng-review, and /plan-design-review (or /autoplan for all three) to lock down product and architecture. (3) Use /design-shotgun and /design-html to iterate on the visual design. (4) Implement the feature yourself. (5) Run /review and /investigate to catch bugs, /qa to test the live behavior. (6) Run /ship to create the PR with full test coverage. (7) Run /land-and-deploy to merge and verify in production. (8) Run /retro to reflect on what shipped. Output: a completely shipped feature with traced reasoning at every step.
Included Skill Groups
Plan with Leadership Perspective8 skills
Strategic product scope decisions from a CEO or founder's viewpoint
Plan Technical Details12 skills
Lock down architecture, data flow, state machines, and test strategy
Design & Iterate Visuals8 skills
Build design systems and iterate on live mockups
Review Code Before Landing12 skills
Find production bugs and completeness gaps in code
Test & Audit Everything24 skills
QA testing, security audits, and performance benchmarks
Ship & Deploy14 skills
Create PRs, merge, deploy, and verify in production
Reflect & Learn7 skills
Weekly retros and cross-session memory
What's the biggest win?
A single builder with gstack ships faster than a traditional team because every step is automated and interconnected. Garry Tan shipped 3 production services and 40+ features in 60 days while running YC full-time, at a 810x productivity gain (241 logical lines/day vs 14 in 2013) because AI handled execution while humans made decisions.
What should I know technically?
Requires Claude Code, Bun 1.0+, Git, and Node.js on Windows only. /browse builds a Chromium daemon (~3s first call, then ~100ms per command) that persists state between skill runs. /design-shotgun uses GPT Image API for variant generation. /cso runs static analysis + LLM threat modeling. All skills run fully offline except image generation. gstack follows the 'Boil the Ocean' principle: when AI marginal cost is near-zero, do the complete thing. Read ETHOS.md for philosophy.
What should I watch out for?
gstack requires Claude Code with proper environment setup (Bun, Node.js on Windows). Some skills like /design-shotgun depend on GPT Image API. Team adoption needs explicit opt-in via CLAUDE.md routing rules or manual skill invocation. Git must be initialized and on a branch for most skills to work.
Key Features
Workflow Methodology
Think → Plan → Design → Build → Review → Test → Ship → Reflect. Each stage is a skill with a specialist role (CEO, Eng Manager, Designer, QA Lead, Release Engineer).
Context Continuity
Every skill writes structured artifacts (design docs, test plans, PR bodies) that downstream skills read automatically. No manual copy-paste between stages.
Live Browser QA
Real Chromium browser via /browse and /qa. The agent has eyes, can test user flows, verify deployments, catch console errors, and generate regression tests automatically.
Multi-AI Review
Code review from Claude via /review and OpenAI Codex via /codex running on the same branch. Cross-model analysis shows findings that overlap (high confidence) and unique catches from each model.
Fully Automated Security & Performance
One command security audits (/cso: OWASP Top 10 + STRIDE), performance benchmarks (/benchmark: Core Web Vitals), and regression test generation (/qa auto-creates tests for every fix).
Tools in this Blueprint
About This Blueprint
- License
- MIT
- Industry
- Technology
- Skills
- 1 workflows, 42 sub-skills, 0 standalone