Prototype
Use when the user wants to build, scaffold, or ship a working MVP from an idea. Covers scope, stack choice, minimal architecture, implementation handoff, validation, and deployment. Trigger on "build an MVP", "prototype this", "scaffold this project", "turn this idea into a working app", "launch this", or /prototype. Do not use when the user only wants to capture, explore, or stress-test an idea; use ideation for that.From its SKILL.md
npx -y skills add kriscard/Skills --skill prototypeAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 13 stars13 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
4.4 KB, 919 tokens by cl100k_base, as published. Nobody here has run it
Prototype Workflow
Structured path from idea → working MVP. Most side projects die in "where do I even start" — this is the scaffolding. Discard it once momentum exists.
Boundary: use ideation when the user wants to think, compare, stress-test, or write planning artifacts. Use this skill when they want to build.
Ask — Calibrate seriousness
Ask first. The answer changes everything.
"Is this a quick side project or something you want to take seriously?"
- Side project / weekend hack: skip formality, move fast, architecture is a one-liner, deployment is optional for now.
- Serious startup / real product: run all phases with full depth.
Cut — Define the smallest usable v1
3 questions max. Don't proceed until all 3 have clear answers.
- Who is this for? (specific person/role, not "everyone")
- What's the one thing it does? (if you need "and", it's two products)
- What does "done" look like for v1? (the minimum that a real user could actually use)
If the user can't answer #2 without "and", help them cut.
Choose — Pick the fastest credible stack
Recommend based on project type and stated skills. Offer 2 options with honest trade-offs — don't just list everything.
Web app (default for solo dev speed):
- Next.js + Vercel — fastest from zero to deployed, great DX, edge-ready
- SvelteKit + Vercel — smaller bundle, less ecosystem, better if React fatigue
API / backend:
- Hono + Bun — fast, minimal, TypeScript-first, great for side projects
- FastAPI — Python, better if the user has ML or data science adjacent work
Mobile:
- Expo (React Native) — ship to iOS + Android from one codebase, Vercel-style DX for mobile
CLI:
- Node.js + commander — if the user knows JS
- Python + click or Typer — if the user knows Python
Database (don't overcomplicate v1):
- Turso (libSQL) or Neon (Postgres) + Drizzle ORM — serverless, free tier
- Supabase — if auth + realtime + storage are needed out of the box
State the trade-off plainly: "Next.js is the default because it ships fastest. Use SvelteKit if bundle size is a constraint or you dislike React's mental model."
Shape — Name the minimal architecture
One-paragraph overview only. No architecture astronautics.
Cover:
- Data model (3–5 entities max for v1)
- Key routes or components
- Auth approach (no auth / magic link / OAuth — pick based on audience)
If the user is over-engineering, say so: "You don't need a message queue for an MVP with 10 users."
Build — Hand off or implement with acceptance criteria
Implementation must include:
- One-sentence product goal
- V1 scope and explicit non-goals
- Stack decision and rationale
- Data model
- Required routes/screens
- Validation commands
- Definition of done
Set up in order: repo → CI if needed → core feature → README. If implementation is delegated or done later, pass the bullets above as the handoff; if implementing now, use them as the acceptance criteria.
Verify — Prove the MVP is shippable
Before declaring done:
- Env vars not committed (
.envin.gitignore) - Domain configured (or Vercel preview URL is fine for MVP)
- Basic error monitoring considered for serious products
- Rate limiting considered for public endpoints
- Analytics considered if user data matters
- README with one-liner description and setup instructions
Skip items that don't apply to a side project, but report what was skipped.
Completion Criteria
Do not declare the prototype complete until:
- V1 scope is 3–5 user-visible capabilities max.
- Stack choice includes one recommended option and one rejected alternative.
- Architecture names the core entities, routes/components, and auth choice.
- Implementation has either been completed or handed off with explicit acceptance criteria.
- Validation commands have run, or failures are reported honestly.
- Final response includes repo path, run command, and deployment URL if deployed.
What ships with it
Read from the repository
Just SKILL.md. No reference files, no scripts.