CodeStable is an AI coding workflow designed for serious engineering projects, addressing the limitations of existing AI frameworks that focus on agent orchestration. It offers a structured, human-in-the-loop approach centered on managing software's lifecycle and its constituent elements—requirements, architecture, features, and bugs—to combat complexity, knowledge loss, and requirement drift. This system is ideal for developers building and maintaining long-term, evolving software where traceability and control are paramount.

How It Works

CodeStable diverges from agent-centric AI coding frameworks by modeling software development around the software's lifecycle and its constituent elements—requirements, architecture, features, and bugs—rather than solely orchestrating agents. This "human-in-the-loop" approach positions programmers as controllers, with AI acting as an efficient executor. The system employs a structured "Entities + Processes" design, where distinct skills (cs-*) guide development phases, enforced by explicit "gates" for reviews and quality checks. This methodology aims to combat complexity, knowledge loss, and requirement drift by ensuring traceability and organization of software artifacts over long-term projects.

Quick Start & Requirements

Installation is straightforward via npx skills add https://github.com/liuzhengdongfortest/CodeStable. Initial setup is performed with the /cs-onboard command. A bash environment is required.

Highlighted Details

• Human-Centric AI: Prioritizes programmer oversight and control, leveraging AI as an execution assistant.
• Software Element Focus: Organizes workflows around core software artifacts (requirements, design, features, issues) for enhanced traceability and long-term management.
• Structured Skill-Based Workflow: Employs a command-driven system (cs-* skills) with explicit quality gates (design review, code review, QA) for disciplined development.
• Knowledge Persistence: Integrates mechanisms (cs-keep, cs-note) for capturing and reusing project-specific knowledge, mitigating information loss.
• AI Branch Protection: Offers an optional hook to enforce AI-generated code to go through worktrees, preventing direct commits to main branches.

Maintenance & Community

The project is authored by [@liuzhengdongfortest]. While specific community links like Discord or Slack are not provided, the roadmap indicates ongoing development, including strengthening the refactoring workflow and adapting to AI model advancements.

Licensing & Compatibility

CodeStable is released under the MIT License, which generally permits broad use, including commercial applications, without significant copyleft restrictions.

Limitations & Caveats

The code refactoring workflow (cs-refactor) is currently in beta. The project's philosophy emphasizes human control, contrasting with fully automated AI agent approaches. Its structured nature may be more suited for complex, long-term projects rather than simple scripts.