Specification-Driven Development (SDD) in 2026: Frameworks, Tools, Formats & Resources

1. Concepts and Foundations

9 entries

What specification-driven development means and where it comes from. Spec-driven development makes the specification the source of truth and generates or checks code against it.

01Spec-driven development with AI (GitHub Blog)GitHub's introduction to Spec Kit and the specify, plan, tasks, implement workflow. 02Understanding Spec-Driven-Development: Kiro, spec-kit, and Tessl (Martin Fowler)A critical comparison of three SDD tools and their levels of spec use. 03Spec-driven development (Thoughtworks Technology Radar)Radar entry on the technique and its trade-offs. 04Design by contractPreconditions, postconditions and invariants as executable specifications. 05Behavior-driven developmentShared, executable behaviour specifications. 06Test-Driven Development (Martin Fowler)Tests as the executable specification. 07Model-driven engineeringGenerating software from models, an older relative of SDD. 08Agentic Software Engineering: Foundational Pillars and a Research RoadmapWhere specifications fit in agent-driven engineering. 09Anthropic: Building Effective AgentsWorkflow patterns that spec-driven pipelines build on.

2. SDD Frameworks and Tools

11 entries

Toolkits that structure requirements, plans and tasks for coding agents. Compare how much of the spec each tool keeps alive after coding.

10GitHub Spec KitOpen-source toolkit and CLI for spec-driven development with many coding agents. 11Kiro SpecsRequirements, design and task documents in the Kiro IDE. 12KiroAgentic IDE built around specs and hooks. 13TesslPlatform exploring spec-centric development. 14OpenSpecLightweight spec workflow that tracks changes as proposals. 15BMAD-METHODAgile multi-agent method with planning and development roles. 16Agent OSStandards and specs layer for AI coding agents. 17Claude Task MasterTurns a product requirements document into tasks for agents. 18AGENTS.mdRepository instructions that coding agents read before working. 19Agent SkillsReusable instruction folders that agents load on demand. 20Cursor RulesPersistent instructions that steer agent behaviour.

3. Specification Formats and Languages

13 entries

The notations in which requirements, contracts and designs are written. Pick formats that tools can validate, not only people.

21OpenAPI SpecificationMachine-readable HTTP API contracts. 22AsyncAPISpecification for event-driven APIs. 23JSON SchemaVocabulary for validating JSON data. 24GraphQL SpecificationThe GraphQL language and execution spec. 25Protocol BuffersTyped interface definition with code generation. 26TypeSpecLanguage for describing APIs that emits OpenAPI, JSON Schema and more. 27SmithyInterface definition language used by AWS. 28Gherkin ReferencePlain-language Given-When-Then scenarios. 29EARS: Easy Approach to Requirements SyntaxTemplates for clear, testable requirements. 30RFC 2119: Key Words for Requirement LevelsMUST, SHOULD and MAY as precise requirement terms. 31Architecture Decision RecordsRecording design decisions next to the code. 32C4 ModelLayered architecture diagrams. 33MermaidDiagrams as text inside specs.

4. Formal and Lightweight Verification

9 entries

Specifications that can be checked mathematically or exhaustively. From model checking to verified code.

34TLA+Specification language for concurrent and distributed systems. 35AlloyLightweight formal modelling and analysis. 36DafnyVerification-aware language with specs in the code. 37LeanProof assistant and programming language. 38Z3 Theorem ProverSMT solver behind many verifiers. 39Frama-CVerification of C programs with ACSL specifications. 40OpenJMLJava Modeling Language tools. 41HypothesisProperty-based testing for Python. 42Software FoundationsTextbook series on verified software.

5. Contract Testing and Conformance

9 entries

Checking that implementations still match the spec. Specs drift unless tests enforce them.

43PactConsumer-driven contract testing. 44SchemathesisProperty-based testing of OpenAPI and GraphQL APIs. 45PrismMock servers and validation from OpenAPI. 46SpectralLinter for OpenAPI and AsyncAPI documents. 47OpenAPI GeneratorGenerate clients, servers and docs from specs. 48BufProtobuf linting, breaking-change detection and generation. 49CucumberRun Gherkin scenarios as tests. 50ReqnrollBDD framework for .NET. 51BehaveBDD framework for Python.

6. Research on Specs and Code Generation

6 entries

Studies of how well models write and use specifications. Early results; evaluate on your own code.

52nl2spec: Interactively Translating Unstructured Natural Language to Temporal LogicsTurning natural language into formal specifications with LLMs. 53Clover: Closed-Loop Verifiable Code GenerationChecking consistency between code, docstrings and annotations. 54Towards AI-Assisted Synthesis of Verified Dafny MethodsLLMs generating verified Dafny code. 55SWE-benchBenchmark of real repository issues for coding agents. 56METR: Early-2025 AI and Experienced Open-Source Developer ProductivityRandomised trial that tempers productivity claims. 57DORA: State of AI-assisted Software DevelopmentResearch on AI adoption and delivery performance.

7. Coding Agents That Consume Specs

6 entries

Agents and protocols used to implement from a spec. Most SDD tools target several of these.

58Claude CodeAnthropic's agentic coding tool. 59OpenAI CodexOpenAI's open-source coding agent CLI. 60GitHub Copilot Coding AgentAssign issues to an agent that opens pull requests. 61Gemini CLIGoogle's open-source terminal agent. 62AiderPair-programming agent in the terminal. 63Model Context ProtocolOpen standard for connecting agents to tools and data.

8. Related Pages on This Site

11 entries

Directories on agents, harnesses and the wider factory. Where the tools around SDD are listed.

64Agentic Software FactoryAgent-driven pipeline from specification to delivery. 65AI Coding AgentsCoding agents and assistants. 66Agentic AI in the SDLCAgents across the software lifecycle. 67Agent HarnessesRuntimes around models. 68AI Agent SkillsReusable agent skills. 69MCP Design PatternsDesigning MCP servers and clients. 70Enabling Tools of AI Coding AgentsTools coding agents rely on. 71Formal MethodsFormal specification and verification resources. 72API Design ToolsTools for designing and documenting APIs. 73Software Testing CoursesTesting learning resources. 74Open-Source Benchmarks for AI Coding AgentsBenchmarks for coding agents.

Frequently Asked Questions

What is specification-driven development (SDD)?

Specification-driven development treats a written specification as the main artifact and source of truth. Requirements, plans and tasks are captured in structured documents, and AI coding agents or generators implement code from them while tests and checks verify the result against the spec. The term is used most often for workflows with AI coding agents.

How is SDD different from TDD and BDD?

Test-driven development writes tests first, and behaviour-driven development writes behaviour examples in plain language such as Gherkin. SDD works one level higher: it specifies intent, design and tasks, and may include tests or scenarios as part of the spec. The approaches combine well, because tests and scenarios keep the spec honest.

Which tools support SDD?

GitHub Spec Kit, Kiro, OpenSpec, BMAD-METHOD, Tessl, Agent OS and Claude Task Master are common starting points. Many work with several coding agents, and repository files such as AGENTS.md and Agent Skills carry standing instructions.

What does an SDD workflow look like?

A typical flow is: write a spec of what and why, clarify open questions, produce a technical plan, break it into tasks, let an agent implement each task, then verify with tests and reviews and update the spec when decisions change. GitHub Spec Kit names the stages specify, plan, tasks and implement.

What formats should specs use?

Use formats that tools can validate: OpenAPI, AsyncAPI, JSON Schema or Protocol Buffers for interfaces, Gherkin for behaviour, EARS or RFC 2119 wording for requirements, and architecture decision records for design choices. Plain Markdown works for intent and context.

What are the limits of SDD?

Specs can drift from code, add review overhead and give false confidence if nobody checks the output. Martin Fowler's analysis notes that tools differ in whether the spec is kept after coding. Contract tests, linting and verification help keep specs and implementations aligned.

How current is this list?

Links were checked in November 2026. Some sites block automated checks and may need a manual click. SDD tooling changes quickly, so check each project for its latest status. This page is a resource list, not engineering advice.