Project Starter

The problem

AI coding assistants forget everything between sessions. Switching tools or starting a new chat means explaining the project again, and decisions, failed attempts and half-finished work get lost.

What I built

Project Starter asks a few questions (name, type, stack, features, rules, and constraints an AI can’t guess) and writes a starter set of project memory files:

• AGENTS.md: the operating manual, with first-run steps, verification, security and architecture rules, and a definition of done.
• CLAUDE.md: a short entry file that points Claude to the same memory.
• PROJECT_BRIEF.md, docs/ (architecture, decisions, security, testing) and logs/ (progress, errors, experiments, handoff).

Presets for web, Android and desktop projects change the verification, architecture and security rules. It never overwrites existing files unless asked.

Where it is used

Wallive’s repository follows this structure, including its decision log and handoff notes, and this portfolio was started with it.