News
GitHub Open Sources Kit for Spec-Driven AI Development
GitHub unveiled an open-source dev kit designed to help developers adopt spec-driven development with AI tools such as GitHub Copilot, Claude Code and Gemini CLI, moving beyond quick "vibe-coding" experiments toward a repeatable, verifiable workflow.
"We treat coding agents like search engines when we should be treating them more like literal-minded pair programmers," the company's Den Delimarsky explains, positioning specifications as living, executable artifacts that guide planning, tasking and implementation.
[Click on image for larger view.] GitHub Spec Kit (source: GitHub).
What It Is
Spec Kit is a GitHub-hosted project (MIT-licensed) that packages templates, a CLI, and prompts to center work around a specification first, then a technical plan, then a set of small, testable tasks that AI agents implement. The repository documents prerequisites, project scaffolding, and ready-to-run commands for popular AI coding agents.
It also lists the core principles of the new offering:
- Specifications as the Lingua Franca: The specification becomes the primary artifact. Code becomes its expression in a particular language and framework. Maintaining software means evolving specifications.
- Executable Specifications: Specifications must be precise, complete, and unambiguous enough to generate working systems. This eliminates the gap between intent and implementation.
- Continuous Refinement: Consistency validation happens continuously, not as a one-time gate. AI analyzes specifications for ambiguity, contradictions, and gaps as an ongoing process.
- Research-Driven Context: Research agents gather critical context throughout the specification process, investigating technical options, performance implications, and organizational constraints.
- Bidirectional Feedback: Production reality informs specification evolution. Metrics, incidents, and operational learnings become inputs for specification refinement.
- Branching for Exploration: Generate multiple implementation approaches from the same specification to explore different optimization targets--performance, maintainability, user experience, cost.
Why It Exists
The Spec Kit post argues that ad-hoc prompting often produces code that compiles but misses intent, or selects stacks you would not choose for production. By contrast, a spec-driven approach makes the specification the shared source of truth that AI tools can execute against and validate at each step. "Specifications become executable, directly generating working implementations rather than just guiding them."
How It Works
The process is organized into four gated phases with explicit checkpoints, and you do not advance until the current phase is validated: Specify, Plan, Tasks, Implement.
- Specify -- describe goals and user journeys; the agent drafts a detailed spec that evolves with feedback.
- Plan -- declare architecture, stack and constraints; the agent proposes a technical plan that respects organizational patterns and standards.
- Tasks -- the agent breaks the work into small, reviewable units that can be implemented and validated in isolation.
- Implement -- the agent tackles tasks while you verify and iterate at each checkpoint.
Getting Started
uvx -- from git+https://github.com/github/spec-kit.git specify init <PROJECT_NAME>
Within your AI agent, the project exposes three steering commands: /specify to generate the specification, /plan to produce the technical plan, and /tasks to derive the actionable task list. The repo also lists prerequisites (Linux/macOS or WSL2, Python 3.11+, uv, Git) and shows how to point the CLI at your preferred agent.
Where It Helps Most
GitHub highlights three scenarios: greenfield projects, feature work in existing systems, and legacy modernization. In each case, the spec captures the stable "what," while the plan and tasks drive the flexible "how," reducing rework and making changes predictable.
For Microsoft-Centric Shops and More
Spec Kit is intentionally tool-agnostic but includes guidance and examples that map cleanly to Microsoft stacks. The repo demonstrates planning that calls out .NET-centric architectures (for example, Blazor front ends and REST APIs) and is designed to steer GitHub Copilot inside Visual Studio Code or other supported environments using the same /specify, /plan, and /tasks prompts.
Availability and Resources
In the announcement developers can watch short demo clips and then clone the GitHub repository to try the workflow with their AI tool of choice.
About the Author
David Ramel is an editor and writer at Converge 360.