Purpose
Guide the user through a structured interview to establish all foundational decisions for a new repository. At the end, produce a reviewed summary and offer to generate config files, instructions, and CI/CD scaffolding.
Platform Adaptation
Before starting the interview, load the reference file for the current agent environment and follow its tool guidance:
Prefer the platform's native structured question/input tool over plain chat whenever one exists. This skill should feel like a guided wizard, not a questionnaire dump.
Defaults
Every question has a marked default (default). If the user says "skip", "default", or gives a blank reply, accept the default and move on. Announce which default was taken so the user can backtrack if needed.
Procedure
Fasttrack — Presets
Before Section 1, ask the user if they want a head start:
"Would you like a preset to skip most questions? I can pre-fill everything for a well-known stack and you only tweak what differs."
Available presets (load from ./examples/presets/):
| Preset | File | What it sets up |
|---|---|---|
| Web + API monorepo (TypeScript) | web-api-monorepo-ts.md |
React 19 + Tailwind v4 + TanStack Query/Router + shadcn/ui, Node.js API, npm workspaces + Turborepo, Vitest, Playwright, ESLint+Prettier, GitHub Actions, PostgreSQL |
| Python API | python-api.md |
FastAPI, uv, pytest, Ruff, GitHub Actions, PostgreSQL |
| TypeScript library / CLI | ts-library.md |
tsup, npm, Vitest, ESLint+Prettier, GitHub Actions npm publish |
| Java Microservices | java-microservices.md |
Spring Boot 3 + Java 21, Maven multi-module, gRPC inter-service, Testcontainers, Helm, GitHub Actions |
| .NET Microservices | dotnet-microservices.md |
ASP.NET Core 9 minimal APIs, MassTransit, EF Core + Postgres, xUnit + Testcontainers, Azure Bicep |
| Python Data Science | python-data-science.md |
Jupyter, pandas/polars, DVC data versioning, uv, Ruff, nbval CI smoke tests |
| Python ML Project | python-ml.md |
DVC pipelines, MLflow tracking, FastAPI serving, uv, Ruff, pytest, Docker |
| Go API Service | go-api.md |
Chi router, sqlc + pgx, golang-migrate, Testcontainers-Go, golangci-lint, GitHub Actions |
| AWS Serverless | aws-serverless.md |
Lambda + API Gateway v2 + DynamoDB, AWS CDK (TypeScript), Turborepo, Vitest, GitHub Actions |
| Azure ML | azure-ml.md |
AzureML SDK v2 pipelines, MLflow, managed endpoints, Azure Bicep IaC, GitHub Actions |
| GCP ML | gcp-ml.md |
Vertex AI Pipelines (KFP v2), BigQuery, Cloud Run serving, Terraform, Workload Identity Federation |
| Start from scratch | (no preset) | Walk through all sections manually |
If a preset is chosen: read the preset file, populate all answers from it, jump directly to Section 5 — Review & Confirm, and show the pre-filled summary. The user only answers questions they want to change.
If "Start from scratch" is chosen: continue with the full interview below.
If the user passes a section name as an argument, jump directly to that section.
At the very start of the interview, tell the user: "Feel free to ask me to explain any term at any time — just say something like 'what is X?' and I'll clarify before we continue. You can also say 'skip' or 'default' on any question to accept the default."
Section 1 — Repo Structure
-
Repo topology — MonoRepo or Polyrepo? (not sure what these mean? just ask)
- Options:
MonoRepo,Polyrepo(default: Polyrepo)
- Options:
-
(Only if MonoRepo) Monorepo tooling — which tool manages the workspace?
- Options:
Nx,Turborepo(default),Bazel,Rush,Lerna,pnpm workspaces (no orchestrator),Other
- Options:
Record answers, then proceed to Section 2.
Section 2 — Tech Stack
Work through each sub-topic below in order. Options for later sub-topics adapt based on earlier answers. User can say "skip" on any sub-topic to accept the default.
2a. Overall Language / Runtime
Options: Java / JVM (Kotlin, Scala, etc.), Python, Node.js / TypeScript (default), .NET (C#, F#), Go, Rust, Other
2b. Build Tooling
Present options driven by 2a:
| 2a answer | Options (default first) |
|---|---|
| Java / JVM | Maven (default), Gradle, Bazel, Other |
| Python | uv (default), Poetry, Hatch, pip + setuptools, Bazel, Other |
| Node.js / TypeScript | npm (default), yarn, pnpm, Bun, Turborepo tasks, Other |
| .NET | MSBuild / dotnet CLI (default), Cake, NUKE, Other |
| Go | go build (built-in) (default), Mage, Task, Other |
| Rust | Cargo (built-in) (default), Other |
| Other | Ask user to describe |
2c. Database
Options (multi-select, default: None): Relational / RDBMS (PostgreSQL, MySQL, SQLite…), Document DB (MongoDB, Firestore…), Key-Value / Cache (Redis, DynamoDB…), Graph DB (Neo4j, Amazon Neptune…), Time-Series (InfluxDB, TimescaleDB…), Vector DB (pgvector, Pinecone, Qdrant…), None (default)
For each selected DB category, ask which specific engine (or "undecided").
2d. API Style
Options (multi-select, default: REST / HTTP): REST / HTTP (default), GraphQL, gRPC / Protobuf, tRPC, WebSockets, None / internal only
2e. Delivery Target
Options: Web (browser) (default), Desktop, Mobile, CLI / daemon, Library / SDK, Multiple
-
If Web: ask about (each skippable):
- CSS approach (default: Tailwind CSS):
Tailwind CSS,Bootstrap / Bulma,CSS Modules,Styled Components / Emotion,Vanilla CSS,Other - State management (default: TanStack Query — server-state only):
TanStack Query (server-state only),Redux / RTK,Zustand,MobX,Jotai / Recoil,None,Other - UI component library (default: shadcn/ui):
shadcn/ui,Radix UI,MUI,Ant Design,Chakra UI,None,Other - Routing (default: TanStack Router):
TanStack Router,React Router,Next.js App Router,None,Other
- CSS approach (default: Tailwind CSS):
-
If Desktop (default: Tauri):
- Shell:
Native (Swift/AppKit, WinUI, GTK),Webview — Electron,Webview — Tauri(default),Webview — Neutralinojs,Other
- Shell:
-
If Mobile (default: React Native):
- Framework:
React Native(default),Flutter,Native (Swift / Kotlin),Capacitor / Ionic,Other
- Framework:
2f. Testing
- Unit test framework (driven by 2a, default noted per runtime):
| Runtime | Options (default first) |
|---|---|
| Java / JVM | JUnit 5 (default), TestNG, Spock, Other |
| Python | pytest (default), unittest, Other |
| Node.js / TypeScript | Vitest (default), Jest, Node test runner, Other |
| .NET | xUnit (default), NUnit, MSTest, Other |
| Go | testing (built-in) (default), Testify, Other |
| Rust | built-in test (default), Other |
- E2E / integration framework (multi-select, default:
None):Playwright,Cypress,Selenium / WebDriver,Puppeteer,Detox (mobile),k6 (load),None(default),Other
2g. Linting & Formatting
Options driven by 2a (default noted per runtime):
| Runtime | Options (default first) |
|---|---|
| Java / JVM | Checkstyle + google-java-format (default), PMD, SpotBugs, Other |
| Python | Ruff (default), Flake8 + isort + Black, Pylint, Other |
| Node.js / TypeScript | ESLint + Prettier (default), Biome, StandardJS, Other |
| .NET | Roslyn analyzers (default), StyleCop, Other |
| Go | golangci-lint + gofmt (default), Other |
| Rust | clippy + rustfmt (built-in) (default), Other |
2h. Cloud & Infrastructure
- Cloud-native? (default: No):
Yes — cloud-first design,Partially — some managed services,No — on-prem / self-hosted(default) - (If Yes or Partially) Cloud providers (multi-select, default:
AWS):AWS(default),Google Cloud,Azure,Cloudflare,Other
Section 3 — AI Tools
3a. AI Coding Assistant
Options (multi-select, default: GitHub Copilot): GitHub Copilot (VS Code agent mode) (default), Claude Code (Anthropic CLI), Cursor, Google Gemini CLI / AI Studio, Codeium / Windsurf, None, Other
3b. AI Development Methodology
-
Does the team follow a structured AI-driven spec methodology? (default: No — ad-hoc)
- Options:
Yes — Spec-Driven Development (SDD),Yes — custom internal methodology,No — ad-hoc(default),Unsure
- Options:
-
(If SDD) Spec format (default: Open Spec):
Spec Kit (structured YAML/MD specs),Open Spec (Markdown-based)(default),Custom format,Other
3c. AI Agent Preferences
These will be recorded in the repo's agent instructions file (e.g. copilot-instructions.md, AGENTS.md, .cursorrules, or equivalent for the chosen assistant).
- Shell preference (default: No preference):
bash,zsh,fish,PowerShell,No preference(default) - Tool use style (default: No preference):
Prefer shell commands over MCP tools when equivalent,Prefer MCP tools,No preference(default) - Confirmation behavior (default: Ask before destructive actions):
Ask before destructive actions(default),Auto-approve safe ops, ask for destructive,Minimal confirmations - Additional free-text preferences (optional, skip to leave blank): e.g. "always use conventional commits", "never modify test files without asking"
Section 4 — CI/CD
4a. CI/CD Platform
Options: GitHub Actions (default), GitLab CI/CD, CircleCI, Jenkins, Azure DevOps, Bitbucket Pipelines, Buildkite, TeamCity, None yet, Other
4b. Infrastructure as Code (IaC)
Options (multi-select, default: None): Terraform / OpenTofu, AWS CDK, Pulumi, AWS CloudFormation, Azure Bicep / ARM, Ansible, Helm (Kubernetes), Docker Compose only, None (default), Other
Section 5 — Review & Confirm
Present the complete summary in a structured format:
## Repo Init Summary
### Repo Structure
- Topology: <MonoRepo / Polyrepo>
- Monorepo tool: <if applicable>
### Tech Stack
- Language / Runtime: ...
- Build tool: ...
- Database(s): ...
- API style: ...
- Delivery target: ...
- [Web] CSS: ... | State: ... | UI lib: ...
- [Desktop] Shell: ...
- Unit testing: ...
- E2E testing: ...
- Linting: ...
- Cloud: ... | Providers: ...
### AI Tools
- Assistant(s): ...
- Methodology: ...
- Shell preference: ...
- Tool use style: ...
- Extra preferences: ...
### CI/CD
- Platform: ...
- IaC: ...
Ask the user to confirm or correct any item before proceeding.
Section 6 — Next Steps
After confirmation, ask the user what they want to do next (multi-select):
- Init the project right now — scaffold the repo layout, config files, and tooling based on the choices above (monorepo config, linter/formatter config, build tool config, etc.)
- Set up AI assistant configs — generate
AGENTS.mdand tool-specific config files (CLAUDE.md,copilot-instructions.md, Cursor rules,GEMINI.md, etc.) using theai-configskill. The assistant choices, shell preference, methodology, and extra conventions from Section 3 will be passed automatically — no re-answering needed. - (Only if SDD was selected in 3b) Install SDD Tools — set up the spec-driven development toolchain and folder structure for the chosen spec format
- Something else — ask the user to describe what they need; proceed accordingly
Execute each selected action one by one. For any action that writes files, describe what will be created and confirm before writing.
Handoff to ai-config skill
When "Set up AI assistant configs" is selected, invoke the ai-config skill and pass the following context from Section 3 so it skips the questions already answered:
tools: <list from 3a, e.g. [copilot, claude]>
shell: <from 3c>
commit_style: <from 3c extra preferences, if mentioned>
confirmation: <from 3c>
extra: <from 3c free-text preferences>
methodology: <from 3b, e.g. spec-driven>
stack:
runtime: <from 2a>
linter: <from 2g>
test: <from 2f unit framework>
Announce to the user which values are being pre-filled before the ai-config skill starts, so they can correct anything without re-running the full wizard.
Guidelines
- One section at a time. Present one section's questions, then wait for the user's answers before moving on.
- Fasttrack first. Always offer presets before starting Section 1. Loading a preset jumps directly to Section 5.
- Skip = default. If the user says "skip", "default", or leaves a reply blank, use the marked default and announce it: "Using default: X".
- Load the current platform reference first. Use the matching file in
./references/before the interview starts so you follow that agent's preferred question UX. - Prefer native question UI. Use the platform's structured question/input tool when available. Fall back to plain chat only if the platform has no dedicated input mechanism.
- One question per turn. Even within a section, keep each prompt focused on a single decision.
- Explain on demand. If the user asks "what is X?" or "explain X" at any point, explain the term concisely, then re-present the current question so they can answer without losing their place.
- Keep the main skill body platform-neutral. Do not hardcode environment-specific tool names here; put those details in the reference files.
- Adapt options to runtime. Build tools, test frameworks, and linters shown must match the selected language.
- Mark branches clearly. Only ask conditional sub-questions when the parent condition is met.
- Always include an "Other" option with a free-text field so users aren't blocked by a missing option.
- Multi-select where applicable. Database, cloud providers, IaC, and AI assistants can have more than one answer. Use native multi-select UI when supported; otherwise collect a comma-separated or repeated-answer list and confirm the parsed values if needed.
- Preserve all answers across sections — the full state is needed for the summary and file generation.
- Be concise in questions. Show a short description per option, not paragraphs.
- SDD state is sticky. If the user selected SDD in Section 3b, always surface the "Install SDD Tools" option in Next Steps.