Claude Code’s pre-flight checklist for new features

What it does This repository is a set of command templates and a folder convention for Claude Code that imposes a structured requirements-gathering ritual. Instead of jumping straight to implementation, the AI must first analyze your codebase, ask you ten yes-or-no questions in two phases, and produce a structured spec with file paths and acceptance criteria. All answers and intermediate findings are written to disk so you can resume across sessions.

The interesting bit The clever part is the forced pacing. The system splits questioning into a high-level phase for product managers and an expert phase after the AI has autonomously dug through your source code, with “idk” accepted as a valid answer that falls back to a claimed smart default. It is essentially a straitjacket for an eager assistant that would otherwise start refactoring on turn two.

Key highlights

• Two-phase interrogation: five high-level yes/no questions, then five expert questions grounded in autonomous code analysis
• File-based state machine: progress, metadata, and generated specs live in a requirements/ directory with numbered markdown files
• /remind command: a panic button for when Claude forgets the rules and starts coding or asks open-ended questions
• Defaults for the indecisive: every question carries a suggested yes/no default based on best practices and codebase patterns
• PM-friendly pitch: marketed as requiring no coding knowledge to answer the initial discovery round

Caveats

• The /remind command exists because the AI is expected to frequently stray off course, start implementing code, or ask open-ended questions instead of sticking to the yes/no format
• The workflow is tightly coupled to Claude Code; there is no indication it works with other assistants or IDEs

Verdict Worth a look if your Claude Code sessions devolve into implementation before you have agreed on what you are building. Skip it if you already enforce a formal spec process outside the chat window, or if you find conversational guardrails more annoying than helpful.