s-kit: A Personal Development Toolkit for Agent Workflows

This article introduces the s-kit project maintained by J03Fr0st, a personal workflow toolkit for AI agent-assisted development. It helps developers turn ideas into executable software features through a standardized design-planning-implementation-verification process. The project supports multiple AI coding assistant platforms, defines a structured agent role and skill system, and includes built-in quality assurance mechanisms. It aims to solve issues like design consistency, contextual coherence, and code quality in AI-assisted development.

With the improvement of large language model capabilities, AI-assisted programming has evolved into an agent collaboration model, but it faces challenges such as aligning design with human intent, maintaining contextual coherence in multi-round iterations, and adhering to code quality standards. The s-kit project was created as a personal workflow toolkit to address these issues.

s-kit defines a structured agent-assisted development process: Brainstorming → Feature Planning → Feature Implementation → Verification & Review → Delivery & Release.

• Brainstorming: Developers collaborate with AI to clarify ideas and explore solutions; optional stress testing is available.
• Feature Planning: Expand into detailed specifications organized in dated folders.
• Feature Implementation: Develop in dependency waves, emphasizing code simplification.
• Verification & Delivery: Pass quality gates (spec check, code review, testing) and finally submit a PR.

Multi-platform Support: Integrates tools like Codex App/CLI, Claude Code, GitHub Copilot CLI, OpenCode, Cursor, Gemini, etc., without binding to a single platform. Agent Roles:

• Analysis category: Codebase Mapper, Pattern Mapper;
• Execution category: Spec Implementer, Code Simplifier;
• Review category: Spec Reviewer, Code Reviewer, Fixer, Security Auditor.

Skill System:

• Core workflow skills: brainstorming, plan-feature, build-feature;
• Supporting skills: grill-me, grill-with-docs, TDD, system debugging, etc. Quality Assurance: npm test performs OpenCode syntax verification, brand cleanup, agent directory checks, and workflow invariant validation (e.g., matching design and spec folders, consistent task statuses, etc.).

s-kit demonstrates the shift of AI-assisted development towards structured collaborative workflows, providing a manageable, reproducible, and auditable model. For individual developers: it helps manage complex projects and ensures design intent is not deviated. For teams: standardized workflows improve collaboration efficiency. It embodies a new paradigm of human-AI collaboration: humans are responsible for creative decisions and quality control, while AI handles detailed execution.

Limitations: Targeted at single-developer scenarios with limited support for multi-person collaboration; the strict workflow adds overhead to small tasks, making it suitable for medium-to-large feature development. Outlook: Enhance team collaboration support, provide domain-specific skill templates, integrate more tool platforms, and explore AI automated testing and document maintenance.