spec-kitty: Govern AI Coding Workflows with Specs

Overview

Spec Kitty turns product intent into repository-managed specifications, plans, and implementation tasks for AI coding agents. It addresses lost requirements and unclear handoffs by keeping mission artifacts and workflow state alongside the code.

Teams can assign work packages to agents in isolated Git worktrees, then track progress through review, acceptance, and merge. The project is local-first, with hosted sync and tracker integrations described as optional.

Key Features

• Guides work from specification and planning through tasks, implementation, review, acceptance, and merge.
• Stores mission artifacts, work packages, and workflow state in the repository.
• Organizes work packages into lifecycle lanes such as planned, in progress, for review, approved, and done.
• Uses isolated Git worktrees to support parallel agent work.
• Provides a next command to identify the next action for an agent and mission.
• Supports multiple coding agents through CLI integrations and slash commands.
• Includes governance and review gates, with mission retrospectives generated by default.

Use Cases

• Product and engineering teams can turn requirements or bug reports into scoped, reviewable work packages.
• Teams running multiple coding agents can isolate parallel changes and track their status in Git.
• Developers can keep acceptance criteria and architecture decisions available across separate AI sessions.
• Organizations establishing repeatable AI-assisted development practices can add explicit review and merge controls.

Getting Started

Install the CLI with pipx:

pipx install spec-kitty-cli

For setup and workflow details, see the README.

Alternatives

• agent-orchestrator: Agent Orchestrator is a desktop workspace centered on supervising agents in isolated Git worktrees, rather than a spec-driven CLI workflow tracked in Git.
• loopx: LoopX focuses on preserving goals, decisions, tasks, and evidence across sessions, rather than managing specification, review, and merge workflows.
• gsd-pi: GSD Pi is a local coding-agent CLI that plans and implements milestone-based work, rather than coordinating a Git-backed spec and review process.
• TaskSync: TaskSync centers on feedback loops and task queues for AI-assisted coding, rather than keeping specs, review decisions, and merge state in Git.

Considerations

• The project describes its default workflow as human-in-the-loop, not a fully autonomous black box. Review and merge decisions remain explicit.
• It is likely overkill for small, one-off edits or projects that do not use Git.
• The README identifies the main branch as a 4.x release-candidate line and says stable launch acceptance is pending. Check the README for current installation guidance.
• The core workflow is local-first; hosted authentication, synchronization, and tracker flows are optional.