Give AI coding agents a structured development process for consistent results
This toolkit reduces the burden of issuing instructions from scratch every time you build features or fix bugs with an AI coding agent. It provides agents with a standardized flow from requirements definition to implementation and verification, ensuring consistent outcomes. You can proceed through complex development stages systematically by installing a Python-based CLI tool and entering specific commands in the agent's chat window.
Three entry points for different situations
Choose 'Spec-Driven Development (SDD)' when creating new features or applications, 'Bug Fixing' when resolving broken existing behavior, and 'Idea Evaluation' when judging the value of an idea. These three are independent entry points used selectively based on the situation, not mandatory steps. SDD is included in the core, while the other two are provided as extensions to install when needed.
Skills executed in the agent chat window
After initializing the project with the specify command in the terminal, you invoke skills in the /speckit-* format in the AI coding agent's chat window. For example, when proceeding with SDD, you execute skills in the order of constitution, specify, plan, tasks, implement, and converge, reviewing the results of each stage before moving to the next. Bug fixing and idea evaluation follow the same method, where you install the relevant extension and then execute the associated skills sequentially. All artifacts are saved as markdown files in the .specify folder within the project to facilitate tracking.
Applying organizational standards and building custom processes
In addition to the default process, it supports various mechanisms to apply organizational standards or create new workflows. Extension adds new commands or quality gates, while Preset adjusts templates or terminology to fit organizational regulations. Workflow automates prompts or shell tasks, and Bundle provides settings tailored to specific roles or teams as a package. You can also search for and install components created by the community through the Catalog, or build an internal-only catalog to restrict usage to approved components.
Installation environment and community source review
Usage requires Python 3.11 or higher, the uv package manager, and a supported AI coding agent. You can install it on Linux, macOS, and Windows environments using the uv tool install specify-cli command. Components provided in the community catalog are not subject to audit or guarantee, so you must review the source code before installing or running them. Additionally, cases where verification is omitted during the bug fixing process are not considered successful, and it is important to confirm that the final verdict result is verified.
