Open Source Framework for Claude Code

You architect.
Agents deliver.

Spec-driven development with self-learning AI agents.

Define your architecture interactively. Create structured specifications. Let autonomous agents build features while you sleep.

AI coding without structure = chaos over time

Sound familiar?

Every feature starts from zero

No architecture guidelines. No conventions. Every time you explain where files go, which patterns apply, how to test.

Inconsistent results

Feature A uses Redux, Feature B uses Zustand. REST here, GraphQL there. The codebase grows — but not together.

Knowledge gets lost

Why did we decide that? Nobody knows anymore. Documentation doesn't exist — or it's outdated.

Meet your AI sparring partner

Design decisions interactively. Execute autonomously.

Specwright guides you through architecture decisions with an AI sparring partner. Every decision is documented. Every feature follows your rules automatically.

0
Workflows
0%
Open Source
LLMs Supported
0+
MCP Tools

Spec in the morning.
PRs by evening.

Specwright UI makes autonomous feature development real.

Auto-Mode

Autonomous Kanban Boards

Toggle Auto-Mode on — all stories on the board are executed autonomously, one by one.

  • Multi-LLM: Assign different models per story
  • Opus for complex architecture decisions
  • Sonnet for standard CRUD operations
  • Local models for sensitive data
Spec Queue

Cross-Project Execution

Queue complete feature specs — not just stories. Works across multiple projects.

  • Feature A from Project X + Feature B from Project Y
  • Press "Start" and walk away
  • Go to sleep, go to the meeting, go to the weekend
  • Come back to finished pull requests

This isn't future tech. This is Specwright today.

From idea to feature in 3 phases

01
/plan-product

Plan your project

Start with an interactive dialog. Your AI sparring partner asks the right questions:

  • → What's your product vision?
  • → Who's your target audience?
  • → Which tech stack fits best?
  • → What architecture pattern makes sense?
Output Documents
product-brief.md
tech-stack.md
architecture-decision.md
roadmap.md
02
/create-spec

Specify your feature

Describe the feature in natural language. Specwright loads your project documents and creates a specification that fits your architecture:

  • → Requirements (what?)
  • → Acceptance criteria (when done?)
  • → Technical refinement (how? where?)
  • → User stories (Gherkin format)
Human-in-the-Loop
Approve requirements
Review implementation plan
Validate story breakdown
Confirm technical details

5-7 checkpoints per specification

03
/execute-tasks

Let agents build

AI agents work through your specification story by story — following your architecture rules:

  • → Skills auto-load (React, Express, etc.)
  • → Architecture conventions enforced
  • → Files land in the right place
  • → Code review at the end
  • → PR gets created automatically
Self-Learning Agents
story-997

Code Review — Full diff review by Opus

story-998

Integration Validation — End-to-end tests

story-999

Finalize — Create PR, cleanup, docs

Everything for structured AI development

Interactive Planning

No prompt engineering. No templates to fill. Your AI sparring partner asks the right questions and guides you through every decision.

Documentation as Code

All decisions stored as structured Markdown. Versioned. Traceable. Readable by humans and AI alike.

Consistency by Design

Every new feature reads your project documents. Architecture, tech stack, coding standards — all enforced automatically.

★ FLAGSHIP

Auto-Mode

Toggle on — all stories execute autonomously. Come back to finished PRs. Spec in the morning, let agents work for hours.

★ FLAGSHIP

Spec Queue

Queue complete feature specs. Cross-project. Press "Start". All get executed — overnight, over the weekend.

★ FLAGSHIP

Multi-LLM Assignment

Assign different models per story: Opus for architecture, Sonnet for CRUD, local models for sensitive data. You decide.

NEW

Persistent Memory

SQLite-based knowledge store with FTS5 full-text search. Your agents remember decisions, patterns, and learnings across sessions.

NEW

Market Validation

Validate product ideas with structured market analysis. Competitive research, landing pages, and ad campaigns — all from one workflow.

NEW

Brainstorming & Feedback

Interactive idea exploration with an AI innovation coach. Transfer results directly into specs, bugs, or product plans.

Where Specwright fits

Specwright is not a code assistant. It's the process layer that makes coding agents predictable.

You
Your Vision & Architecture Decisions
SPECWRIGHT
Planning, Specifications & Orchestration
Interactive sparring, structured specs, Auto-Mode, Spec Queue
Execution Engine
Claude Code
Autonomous coding agents via integrated terminals
Your Codebase
Consistent, documented, architecture-compliant code

What changes with Specwright

Without Specwright
  • Every feature starts from a blank prompt
  • No architecture guardrails for agents
  • Manual prompting, hoping for consistency
  • Decisions undocumented, context lost
  • One agent, one task, one terminal
With Specwright
  • Every feature follows your architecture
  • Structured specs enforce your conventions
  • Interactive sparring partner instead of guessing
  • Every decision documented as structured Markdown
  • Auto-Mode: queue specs, agents build autonomously

Specwright makes Claude Code better — by giving it structure, context, and your architectural rules.

100% Open Source. No Strings Attached.

The complete framework is free forever. You only pay for LLM API usage.

Framework
Free Forever

Everything you need to get started

  • All 37 workflows
  • Full documentation
  • Multi-LLM support
  • Auto-Mode & Spec Queue
  • Persistent Memory DB
  • Market Validation workflows
  • Community support (GitHub)
  • MIT License — use it anywhere
Get Started on GitHub
Kickstart Session
$199

90 minutes to mastery

  • 1:1 video call with the creator
  • Setup Specwright in your project
  • Best practices for your tech stack
  • Workflow customization tips
  • Q&A — bring your questions
Book Your Session

No subscriptions. No hidden fees. Just open source software.

Frequently Asked Questions

Specwright is an open-source framework for Claude Code CLI. It consists of 34 structured workflows that guide you through project planning, feature specification, and implementation — with AI as your sparring partner, not autopilot.

Copilot and Cursor are code assistants — they help you write code. Specwright is a process framework — it helps you think. You work out architecture decisions interactively, document them, and every feature automatically follows your rules.

Yes. The /analyze-product workflow analyzes your existing codebase and creates project documents retroactively. Zero migration required — Specwright adapts to you.

Specwright is LLM-agnostic. It works with Anthropic Claude, OpenAI, local models via Ollama, and any OpenAI-compatible API. Mix and match based on task requirements and cost.

The framework itself is 100% open source under MIT license. You only pay for LLM API usage (Anthropic, OpenAI, etc.). Enterprise support with SLA is optionally available.

No. Specwright guides you through interactive dialogs. The AI sparring partner asks the right questions — you just answer and decide.

Ready to architect smarter?

Start your next project with an AI sparring partner that understands and respects your architecture.