Spec Kit is an open-source toolkit released by GitHub, designed to help developers use AI coding agents with Spec-Driven Development (SDD) or other custom workflows. It provides AI coding agents with structured workflows, reusable templates, and documented outputs.
Purpose
The core purpose of Spec Kit is to make AI coding agents work according to explicit specifications rather than generating code directly. It applies to three independent entry points: Spec-Driven Development (SDD), bug fixing, and idea assessment. These three entry points are not mandatory phases; SDD is built into the core, while bug fixing and assessment are optional extensions that are installed when needed.
Main capabilities
- Spec-Driven Development (SDD): Turns requirements into specifications, technical plans, and executable tasks, then guides implementation. The constitution is executed once per project, and each feature proceeds through the specify → plan → tasks → implement → converge workflow. Implement → converge can be repeated until the convergence report shows Converged. Clarification, checklists, and consistency analysis can also be added as additional quality gates.
- Bug fixing: Separates diagnosis, fixing, and verification, allowing the agent to fix the assessed cause and check the original symptoms. There is no need to first run the SDD feature workflow. After installing the optional extension via the CLI, invoke the assess → fix → test skills in sequence in the agent chat. Reports are stored in the .specify/bugs/login-crash/ directory, with a final verdict of verified, partial, or failed. A lack of verification does not count as a successful fix.
- Idea assessment: Gathers evidence before committing to an idea, whether or not it ultimately becomes software. This independent workflow can even run in projects with no source code. Invoke the intake → research → define → shape → decide skills in sequence; outputs are stored in the .specify/assessments/offline-mode/ directory, with a final decision of go / needs-clarification / kill. Unknown issues can be resolved by directly refining existing Markdown outputs or with the help of the agent, rather than regenerating the entire phase. A go decision can be handed off to /speckit-specify continues to build; stopping with a documented reason is also a useful outcome.
- Custom or bring-your-own process: extensions add capabilities, presets adjust existing behavior, workflows automate steps, and bundles package role-based settings. Project-local overrides can be used for one-off template modifications.
Usage
Requires Python 3.11+, uv, and a supported AI coding agent, and supports Linux, macOS, or Windows. For CLI-only setup, run installing Spec Kit in a terminal and create a project. The examples use GitHub Copilot's default skill mode; replace copilot with another integration key to use another supported agent. For existing code, refer to the existing project guide. For pinned versions, other installers, CI, or troubleshooting, see the installation documentation. To update an existing installation, see the upgrade documentation.
After installation, start the coding agent in the project directory and choose one of the above processes. In the agent chat, invoke each /speckit-* skill one by one and review the results before continuing. These are agent skills, not terminal commands. Other agents and modes may use different invocation syntax.
When to use it
Spec Kit is suitable for developers who want to use AI coding agents in a structured, reviewable way, including: building new features from scratch, fixing bugs in existing code, evaluating whether a new idea is worth the investment, and teams that need to turn specifications into executable plans.
Limitations
Requires Python 3.11+, uv, and a supported AI coding agent. Skill invocation syntax may differ across agents and modes. Bug fixes and idea evaluation require additional extensions to be installed. A fix without verification is not considered successful. For the specific list of supported agents, installation details, and troubleshooting information, see the official documentation.
