A 65-section project planning framework that turns a product idea into fully-specified architecture, tasks, and sprint plans — ready to build. 100+ autopilot steps. 13 phases. One command.
The Kit doesn't write code. It produces the blueprints. Service specs, screen specs, API contracts, task files, sprint plans, design systems, testing strategies, and deployment configs. When it's time to build, the GSD execution engine reads those specs and turns them into working software.
Drop the Kit into your project directory and run /kit in Claude Code. The orchestrator detects your project's maturity — greenfield, existing codebase, or pivot — and routes you to the right path automatically.
From there, it's a guided process. The Kit asks questions, you answer. It generates specs, you approve. At every major checkpoint (called "gates"), it pauses for your sign-off before continuing.
Domains and categories compose additively — a healthcare marketplace gets both the healthcare domain pack and the marketplace category pack. Rules stack, questions merge, entities combine. No artificial branching that forces you to choose.
Each domain pack includes specialized intake questions, domain personas, business rules, regulatory context, entity definitions, and tribunal research rounds. Each category pack adds product-shape patterns, architecture decisions, and UX conventions.
Domains
Categories
Intersections are real. An automotive SaaS marketplace gets all three packs — automotive domain rules, SaaS multi-tenancy patterns, and marketplace two-sided economics. The Kit composes them, deduplicates overlapping questions, merges personas, and produces a unified specification that respects all three contexts.
dev_docs/ directory with everything needed to start building — or hand off to a team.One per backend service. Business rules, API endpoints, data models, error handling, auth requirements. Scored for depth — minimum 2,000 words, 10+ business rules, 8+ endpoints per spec.
ArchitectureEvery screen in the application with component trees, state shapes, interactions, responsive behavior, and accessibility requirements. Includes domain-specific design patterns.
DesignFull endpoint documentation with request/response shapes, validation rules, error codes, rate limits, and authentication. Generated from service specs and cross-validated.
ArchitectureConcrete development tasks with acceptance criteria, file plans, test requirements, effort estimates, and 8-layer coverage checklists. Grouped into integration units and sequenced into sprints.
PlanningDesign direction document with color tokens, typography scale, spacing system, layout patterns, animation tokens, accessibility requirements, responsive breakpoints, and interaction patterns.
DesignEnd-to-end workflow specifications with full screen state, component trees, API calls, UX reasoning, and every decision branch. Each workflow includes a matching Playwright E2E test.
Design60–100 research files from a 10-round adversarial process covering competitors, personas, technical feasibility, market analysis, and domain-specific research rounds.
Research17 enforcement gates, proof artifact collection, design compliance verification, responsive screenshot capture, and Definition of Done validators. Mechanically enforced — not documentation.
QualityMarketing plans, legal documents (privacy policy, ToS, DPA), financial models, BI dashboards, SEO strategy, support platform setup, and incident response runbooks.
Operations| Section | What It Covers | Key Templates |
|---|---|---|
| Discovery & Intake | Project brief, domain classification, intake questions, creative expansion | Project brief, domain classifier, 54 domain/category packs |
| Tribunal Research | 10-round adversarial research + 6 domain-conditional rounds | Competitive analysis, persona cards, feasibility verdicts |
| Architecture | Database design, API design, state management, component library | DB schema, migration plan, component library, API patterns |
| Documentation | Service specs, screen specs, design docs, state files, contracts | Service spec, screen spec, integration unit, design direction |
| Phase Planning | Phase templates, task templates, sprint planning | Task file, phase plan, timeline, dependency map |
| Dev Infrastructure | Docker, monorepo, linting, hooks, CI/CD, Claude Code config | Docker compose, turbo config, eslint, pre-commit hooks |
| Development Workflow | Session rituals, onboarding, verification loops, GSD protocol | Session ritual, verification macros, GSD verification protocol |
| UI Design System | Design tokens, components, anti-slop rules, domain patterns | Design tokens, 27-domain design pattern library |
| Quality & Testing | Test configs, enforcement gates, design compliance, proof artifacts | 17 enforcement gates, 8-layer task verification, DoD validator |
| Generators | 54 auto-generation prompts with depth requirements | Service hub, screen catalog, Context Bible, battle cards |
| Mobile | Platform selection, mobile UI, native features, mobile deployment | Mobile architecture, push notifications, offline-first |
| Marketing & Growth | SEO, content, email, social, launch planning, PLG | Launch checklist, content calendar, growth loops |
| Legal & Privacy | Privacy policy, ToS, DPA, EULA, GDPR, data flow mapping | Privacy policy, cookie policy, consent management |
| Financial | Revenue projections, unit economics, runway, investor docs | Financial model, pitch deck, cap table, term sheets |
| Operations | Incidents, support, billing, BI, dashboards, experiments | Runbooks, SLA templates, dashboard specs, A/B framework |
| Hardening | 17 rounds of audit, enhancement, depth verification, expansion | Audit checklists, enhancement logs, expansion templates |
The GSD engine reads task files and builds features sprint by sprint — 20–40 tasks per session. Each task goes through an 8-gate verification sequence before it can be marked complete: type check, tests, lint, build, design compliance, screenshots, state completeness, and integration verification.
Failed tasks get 3 retry attempts with automated fixing. After 3 failures, the task is escalated to BLOCKED status with a full diagnostic report.
After shipping, the Kit scaffolds monitoring (Sentry, analytics, uptime), feedback collection (widgets, NPS, knowledge base), and a signal-driven iteration loop. Bugs trigger automatic fix-test-deploy cycles. User feedback aggregates into the next sprint plan.
The lifecycle engine runs as long as you want it to — or you can detach and manage the project manually at any point.
Starting from scratch. Full 100+ step process from discovery through build. Every section applies. The Kit produces the complete specification and lets you build with full confidence.
Existing codebase that needs improvement. The Kit audits what you have, scores it, identifies gaps, and generates tasks to bring the project up to spec. Skips discovery, goes straight to audit.
Pivoting or forking an existing application to a new vertical. The Kit maps what can be reused, what needs to be rebuilt, and produces a migration plan with the new domain's requirements layered in.
| Gate Mode | How It Works | Best For |
|---|---|---|
| Manual | Every gate pauses. You review and approve each one. | First project, regulated domains, learning the Kit |
| Semi-Auto | Structural gates pause, informational gates auto-pass. | Experienced users, standard projects |
| Auto | Most gates auto-pass. Only major milestones pause. | Repeat projects, tight timelines |
| Autopilot | Full autonomous execution. Pauses only at session boundaries. | Rebuilding a well-understood product |
Domain-adaptive gates. Gate criteria aren't one-size-fits-all. Healthcare projects get HIPAA compliance checks. Fintech projects get PCI-DSS verification. Marketplace projects get dual-side coverage validation. The criteria stack based on your project's domain and category classification.