agentic-coding

This skill set addresses the lack of a unified process and standards in AI-assisted coding. Developers often lose context when switching between multiple tasks, leading to inconsistent code quality and inefficient reviews. The framework connects requirements, design, task breakdown, implementation, and review into a closed loop through Specification-Driven Development.

Usage is straightforward: simply integrate this repository into your project and follow the provided skill workflows. Start by creating a specification with starwave:creating-spec, then use next-task or make-it-so for incremental implementation, and finally format commits with the commit skill. The entire process is automated via Claude Code's built-in agent mechanism—you only need to describe requirements in natural language.

It is well-suited for developers who require high-quality code output and team collaboration, especially individuals or teams already using Claude Code and looking to boost development efficiency. For project managers and technical leads responsible for code review and architecture design, skills like peer-review-validator and local-review can significantly reduce review overhead.

We recommend adopting this framework consistently when starting new feature or module development. Note that it is primarily designed for the Claude Code environment; some skills rely on external AI systems (e.g., Gemini/Codex) but default to using local sub-agents on personal machines to prevent data leakage. We advise reading spec-workflow.md first to understand the complete workflow.