AI Project Starter

A portable template repo for hardening a software project idea before implementation.

The workflow starts from a rough idea, generates project.md, then runs gated AI passes for architecture, implementation, planning, data security, and final risk review. Prompts are plain Markdown so they can work in OpenCode, Claude Code, Cursor, ChatGPT, or similar tools.

Quick Start

1. Create a new repo from this template.
2. Run npm run init:project.
3. Tell your AI agent: Run /project-brief using my project idea: <your idea>.
4. Continue through the commands in order.
5. Run npm run check:gates to verify the next blocked gate.

Workflow

Shared rules live in .ai/workflow.md.

Commands live in .ai/commands:

• /project-brief -> project.md
• /architecture -> architecture.md
• /implementation -> implementation.md
• /planning -> plan.md
• /security -> security.md
• /review -> final readiness review

If your AI environment does not support slash commands directly, paste or reference the matching Markdown file.

Project Files

• project.md product intent and constraints
• architecture.md CTO-level decisions
• implementation.md implementation decisions
• plan.md sequencing, checkpoints, verification, and open decisions
• security.md data security, privacy, abuse, and operational safety decisions
• future-concerns.md deferred risks
• manifest.local.md local workflow state, gitignored

Core Rules

• Strict gates: each phase must pass before the next starts.
• Multiple choice first: every major decision includes Need more info.
• No silent assumptions: ambiguous answers require a follow-up before recording.
• Durable context: decisions are written to Markdown source-of-truth files.
• Light automation: scripts scaffold and validate files, but never make product decisions.