skills: Grilling sessions to reduce agent chatter and align requirements

It focuses on solving communication errors and code quality degradation that occur when using AI agents like Claude Code or Codex. Instead of existing process tools like GSD or BMAD taking control, it is provided as small, composable skills that users can directly modify and extend. Based on decades of engineering experience, these skills work regardless of the model.

To solve the problem of agents not accurately understanding requirements, it provides the /grill-me and /grill-with-docs skills. These perform a 'grilling session' where the agent asks the user detailed questions to align requirements before development begins. Additionally, by organizing the project's domain terms into a shared language and automatically generating CONTEXT.md and ADRs, it reduces unnecessary agent chatter and improves consistency in code naming.

For code quality and architecture management, the /tdd and /improve-codebase-architecture skills are key. /tdd enforces the red-green-refactor cycle, prompting agents to write failing tests first, while /improve-codebase-architecture scans the codebase and presents opportunities to deepen modules in a visual HTML report. This process is not just simple structural improvement, but a systematic approach to controlling software entropy generated by agents.

Skills are divided into orchestration skills called directly by users and reusable discipline skills called automatically by agents. It integrates with various issue trackers such as GitHub and Linear, and depending on the installation method, you can subscribe as read-only bundles or copy them as editable files within the project. It is suitable for developers who want to maintain productivity while adhering to fundamental software engineering principles in AI-assisted development.