SDD: A Spec-Driven Development Framework for Claude Code

SDD (Spec-Driven Development) is an open-source spec-driven development framework designed exclusively for Claude Code, released by genkovich on GitHub in June 2026. Its core innovation lies in transforming traditional imperative AI programming into conversational collaboration. Through 12 atomic Socratic skills and a TDD implementation engine, it supports agent team mode and dynamic workflow mode, with specs at its core to enhance the work quality of AI programming assistants.

Original project link: https://github.com/genkovich/sdd

SDD was proposed to address the limitations of Test-Driven Development (TDD): TDD prioritizes tests but may miss the deep understanding of the 'why' and 'what' of requirements. Building on this, SDD emphasizes 'spec-first'—establishing a complete shared understanding of requirements through structured spec definitions.

Additionally, SDD draws on the Socratic method of questioning, using continuous inquiries (e.g., "Why is this feature needed?" "Under what conditions will it fail?" "What are the success criteria?") to reveal deep-seated assumptions and potential issues, helping to clarify requirements and boundaries.

12 Atomic Socratic Skills

The SDD framework defines 12 composable atomic skills, divided into three categories:

• Requirement Clarification Skills: Intent Recognition, Boundary Exploration, Dependency Analysis, Constraint Extraction
• Design Deduction Skills: Concept Modeling, Interface Design, State Analysis, Invariant Definition
• Implementation Guidance Skills: Step Decomposition, Verification Strategy, Refactoring Identification, Documentation Generation

TDD Implementation Engine

SDD includes a dedicated TDD engine that supports two modes:

• Agent Team Mode: Claude Code is assigned roles such as spec engineer, test engineer, and implementation engineer to simulate real-team collaboration
• Dynamic Workflow Mode: Automatically adjusts workflows based on task types (exploratory, implementation, maintenance, optimization)

SDD promotes the evolution of AI-assisted programming from 'code generation' to 'collaborative development', addressing several pain points in AI programming:

1. Requirement Understanding Bias: Reduces understanding errors through Socratic dialogue
2. Missing Boundary Cases: Discovers edge cases via systematic boundary exploration
3. Technical Debt Accumulation: TDD practices ensure code is testable and maintainable
4. Context Loss: Specs as shared context reduce information decay

Claude Code's strong reasoning capabilities align well with SDD's conversational interaction mode, fully leveraging its advantages.

SDD applies to various development scenarios:

• Complex Feature Development: Establishes a clear requirement baseline, coordinates multi-file and complex business logic
• Legacy Code Refactoring: Reverse-engineers specs to safely refactor existing code
• Team Collaboration: Uses specs as a communication medium to ensure AI-generated code meets team standards
• Learning and Teaching: Socratic methods help understand the 'why' rather than just the 'how'

SDD is an important methodological framework in the field of AI-assisted programming. Its core value lies in emphasizing that 'understanding the problem' is more important than 'writing code'. Through 12 atomic skills and a TDD engine, it provides Claude Code users with a systematic and reliable development method.

In the future, as AI programming assistants become more widespread, such methodologies will help developers establish more effective collaboration models with AI, combining human domain knowledge with AI's code generation capabilities to improve development efficiency and quality.