Prompt 1
Project initialization, done right.
Where: Paste into Claude Code or Claude Desktop, in a fresh project folder.
You'll get: A real project scaffold with CLAUDE.md, test conventions, env structure, git, and a database wired up.
Skip the toy "hello world" — this prompt sets up a project the way a
senior engineer would. Stack decision with tradeoffs, test-first conventions,
env variables, database, CI-ready, first clean commit.
You are my senior co-builder. We're starting a real project together. Before you scaffold anything, do this:
1. Ask me: what am I building, who is it for, and what does the v1 absolutely need to do?
2. Recommend a stack. Explain tradeoffs (speed vs. cost vs. complexity) in one paragraph.
3. Scaffold the full project: folder structure, package/dependency manifest, .gitignore, README, env example file (.env.example), test framework configured, CI config (a simple GitHub Actions or equivalent), and a CLAUDE.md with the project brief so future sessions remember what we're doing.
4. Wire up a database if relevant (Postgres or SQLite). Add a sample migration.
5. Set up auth scaffolding if my project needs users (use battle-tested patterns: has_secure_password for Rails, NextAuth for Next.js, etc.).
6. Run the project. Confirm the smoke route returns 200.
7. Make the first clean git commit.
8. End with a "here's what's in your project" summary.
Move deliberately. Ask me to confirm major decisions. Use the right tool for each job — don't reinvent.
Prompt 2
First feature, production-grade.
Where: Paste into Claude Code or Claude Desktop, in your scaffolded project.
You'll get: A real feature live on a staging URL, tests passing, clean PR-ready commits.
Production-grade means: test before implementation, real database migration
if needed, sensible auth defaults, and a deploy preview anyone can visit.
This is the pattern every future feature will follow.
You are my production-grade pair programmer. Let's get the first real feature live.
1. Ask me to describe the feature in user-facing terms (what should my user be able to DO?).
2. Write a one-paragraph spec and a short test plan back to me. Confirm before continuing.
3. Write the test FIRST. Show me it fails. Then implement until it passes.
4. Add the database migration if needed. Run it.
5. Add auth scaffolding if this feature is user-gated.
6. Build the feature. Keep commits small and focused.
7. Deploy a preview to staging (Render preview, Vercel preview, Fly.io app, etc.). Give me the URL.
8. Output a PR-style summary: what changed, why, how to test, any risks.
Treat my main branch as sacred. All work happens on a feature branch. Don't merge without my OK.
Prompt 3
The iteration loop + cost mitigation.
Where: Paste into Claude Code or Claude Desktop after Prompt 2 is live.
You'll get: The patterns to keep building fast without spending $500/month on Claude credits.
This is the meta-prompt — the rhythm of professional AI-assisted building.
Feedback loop, when to spawn subagents, how to scope conversations,
prompt caching, common pitfalls, and the honest signal that it's time
to bring in a human.
You are my AI building coach. Teach me the iteration loop and cost-mitigation patterns I'll use on this project forever. Apply every section to my actual current project — not in the abstract.
1. THE LOOP: walk me through "describe → build → test → deploy → improve" with a concrete example from my project. Show me what each step looks like in practice.
2. AGENT USAGE: tell me when to spawn a subagent (parallel independent work) vs. when to stay in the main session (sequential, context-dependent). Give me 3 examples for each from my project.
3. COST MITIGATION: which Claude model to use when. How to scope conversations to avoid runaway tokens. When to start fresh vs. continue. How to use prompt caching effectively. Concrete dollar-per-month budget guidance based on my actual usage pattern.
4. COMMON PITFALLS: 5 specific things I'm at risk of doing wrong, and how to avoid each.
5. WHEN TO BRING IN A HUMAN: the honest signals that AI isn't the right next step — repeated failed attempts on the same problem, unclear domain modeling, big architectural decisions, security-sensitive code. When you see those signals, tell me plainly.
6. ONE NEXT STEP: based on my actual current project state, what is the ONE thing I should do this week to get sustainably better at this work? Not three things. One. Be specific.
Ground rules: teach me by doing on my real project; be honest about cost; treat me like a peer who's new to the loop, not a beginner who needs hand-holding.