ai-project-bootstrap
Docs / Getting started / The wizard, question by question

The wizard, question by question

Seventeen questions at most, asked in a fixed order, with anything irrelevant skipped automatically. In practice a preset answers most of them for you.

Questions come from config/categories.json and are asked in its order field. A question whose showWhen condition isn't satisfied by earlier answers is never shown, and a category with no installed modules is skipped entirely — so the catalogue can grow without changing the questions.

The two gating questions

These two shape the rest of the wizard rather than selecting a technology. Their answers never reach the module resolver.

Question
Type
Effect
Which AI coding tools do you use?
pick any
Decides which rule formats get written. Defaults to Cursor + Claude Code.
What are you building?
pick one
Mobile, Web or Both. Choosing Both asks for a mobile platform and a web framework.

The eight AI tool choices

Cursor — .cursor/rules/<id>.mdc
Claude Code — .claude/skills/<id>/SKILL.md, frontmatter synthesised from the manifest
GitHub Copilot — .github/copilot-instructions.md plus .github/instructions/<id>.instructions.md
Continue.dev — .continue/rules/<id>.md
Cline — .clinerules/<id>.md
Roo Code — .roo/rules/<id>.md
OpenAI Codex — reads the generated AGENTS.md, no per-technology directory
Gemini CLI — reads the generated GEMINI.md, no per-technology directory

The stack questions, in order

Order
Category
Type
10 / 15
Mobile platform · Web framework
pick one — shown by target
20
Backend
pick one
30
Authentication
pick one
40
Database
pick one
45
Cross-cutting features
pick any
50
Payments
pick one
60
Analytics
pick any
70
Crash reporting
pick any
80
Notifications
pick one
90
Storage
pick one
100
Email
pick one
110
Monitoring
pick one
120
CI/CD
pick one
130
Testing
pick any
140
Deployment
pick one
Every stack question is optional

Only 'What are you building?' is required. Everything else accepts None, and a category you skip simply contributes nothing — no placeholder, no TODO.

Before anything is written

The wizard shows what a preset pre-filled and lets you back out to a fully custom run. It then prints the resolved file list and a one-line cost estimate. --dry-run stops there permanently.