Living specs. Autonomous builds.
gspec gives your AI coding tools the structured product context they need — then an agent team that can take an idea all the way to working code.
An autonomous build pipeline, a self-improving memory, and deterministic quality gates — across six AI coding platforms.
The problem with AI-assisted development
AI tools lack project context
Your AI coding tools don't know your product vision, design system, tech stack decisions, or team standards. Every prompt starts from zero, producing inconsistent code that doesn't match your project.
Specs drift from reality
Documentation written at kickoff becomes stale within weeks. Your AI tools reference outdated specs, or worse, get no context at all — leading to decisions that contradict your evolving codebase.
From idea to built product — autonomously
gspec is no longer just a spec generator. Its autonomous build takes your idea and drives the whole workflow deterministically — with quality gates that catch their own mistakes and a run you can resume.
One command
Run /gspec-build, answer a short intake, and the pipeline drives specs → plan → code on its own.
Nine gated stages
Each stage is an isolated agent checked by an independent validator — the producer never grades its own work. A human review gate sits between the specs and the code.
Builds that verify
A generated verify.sh builds and tests every module. The run pauses rather than shipping something broken.
A structured workflow for AI context
Define your product once, then let AI build with full context. Only Define and Build are required — everything else adds optional depth.
Define
RequiredProduct identity, design, stack, standards
profile style stack practices Research
Competitive analysis & gaps
research Specify
Feature PRDs
feature Architect
Technical blueprint
architect Plan
Ordered plan
plan Analyze
Cross-spec consistency or per-feature ambiguity sweep
analyze Build
RequiredSpec-driven implementation
implement Specs stay in sync with your code through automated spec-sync — update once, context updates everywhere.
Sixteen commands, backed by an agent team
Each command adopts a professional perspective and delegates the real work to specialized agents behind a quality gate. See how it works →
Define
Establish your product foundation
profile Product Strategist Define what your product is, who it serves, and why it exists
style UI Designer Generate a renderable HTML design system (style.html) or Markdown style guide (style.md), plus support for design-tool mockups in gspec/design/
stack Solutions Architect Define your technology choices, frameworks, and infrastructure
practices Engineering Lead Set development standards, testing conventions, and workflows
Optional Depth
Add precision where it matters
research Market Analyst Analyze competitors and identify feature gaps
feature Product Manager Write one or more PRDs with prioritized capabilities and acceptance criteria
architect Technical Architect Design the technical blueprint — data model, APIs, project structure
plan Engineering Lead Decompose a feature PRD into an ordered, dependency-aware plan with parallel-execution markers
analyze Quality Analyst Cross-reference all specs and resolve contradictions, or scrutinize a single PRD for ambiguity
audit Specification Auditor Check specs against the actual codebase and update them to match reality
Build & beyond
Implement, verify, and improve
build Build Orchestrator New Autonomously drive the whole workflow — from an idea to working, verified code
implement Senior Developer Read all specs and build software with complete project context
qa QA Lead New Run the producer ≠ checker quality gate over your specs and code on demand
teach Steward Correct an agent now, without waiting for it to get something wrong first
memorize Steward New Review what runs recorded and commit the worthy memories into your skills
migrate Migration Specialist Upgrade existing gspec files to the current format
Design-tool native
AI design tools like Figma, v0, and Framer AI produce real, renderable output. gspec meets them where they are — with an HTML design system your AI agent can render, and a design folder that captures mockups as first-class implementation context.
style.html Renderable design system
A single self-contained HTML file with your design tokens as CSS variables, live color swatches, typography specimens, and real styled component previews. Open it in a browser. Inspect it with a design-aware AI. Markdown style guides still work — pick whichever format fits your workflow.
/gspec-style gspec/design/ Mockups as context
Drop HTML, SVG, PNG, or JPG mockups exported from Figma, v0, Framer AI, or any other design tool into the design/ folder. When /gspec-implement builds UI, it reads the relevant mockups and matches their layout and visual hierarchy within the style guide's token constraints.
drop files in gspec/design/ Ship more products, faster — on your terms
Every product you build refines your playbook. Save the specs that represent your standards — your stack, your style, your practices — and reuse them to launch new projects that meet your requirements from the start.
No more re-explaining your preferences to an AI. No more starting from scratch. Just your proven blueprints, ready to go.
Define once
Build your ideal stack, style, practices, and features for a project — exactly the way you want them.
Save what works
Save specs to your personal library. Bundle them into playbooks that capture your complete project blueprint.
gspec save Launch the next one
Seed a new project from your playbook in seconds. Same standards, same quality — without repeating yourself.
gspec restore playbook/my-stack One install, every platform
gspec works with the AI coding tools you already use. Same specs, consistent context, everywhere.
Support isn’t identical across these tools. Compare platform capabilities →
Ready to give your AI tools context?
One command installs structured specification skills into your AI coding tool.
Then follow the Getting Started guide to define your first spec.