Prompts That Ship
Real prompts from real projects. These aren't hypotheticals—each one produced working software. Learn the patterns that turn natural language into production code.
The Art of Effective Prompts
Great prompts share common traits: they're specific about outcomes, clear about constraints, and explicit about completion criteria.
Below are prompts I've actually used to build software. Each includes the prompt itself, the context in which I used it, and the outcome it produced.
Key Principle:
Don't tell Claude how to code. Tell Claude what you want to exist when it's done. Let the AI figure out the implementation—that's what it's good at.
1. Portfolio Site Prompt
"Create interactive web site hosted on cloudflare pages, with cloudflare workers ai integration. Terminal/coder aesthetic. Floating AI chat. Projects organized by verticals..."
Context
Needed a portfolio site that showcased vibe coding methodology. Wanted it to feel like a terminal to reinforce the developer brand.
Result
This entire tasteful-vibes site—built in a single Claude Code session. Astro, Tailwind, Cloudflare Workers AI integration, all working.
Why It Works:
Specifies the deployment target (Cloudflare Pages), the aesthetic (terminal/coder), key features (floating AI chat), and organization principle (verticals). Gives Claude enough direction without micromanaging implementation.
2. Repo Organization Prompt
"Git submodule all repos since September 2025. Calculate file counts and sizes. Rank by total file size."
Context
Had 100+ repos created during an intense vibe coding period. Needed to understand the scope and organize them meaningfully.
Result
100+ repos organized automatically. Complete with file counts, size rankings, and submodule structure for the portfolio site's project grid.
Why It Works:
Clear time boundary (September 2025), specific operations (submodule, calculate, rank), and explicit sorting criteria (total file size). No ambiguity about what "organized" means.
3. Preference Gathering Prompt
"Ask me many multiple choice questions to help you understand my taste and help you proceed."
Context
Starting a design project where I knew what I wanted but couldn't articulate it. Needed Claude to extract my preferences through structured questions.
Result
Claude used its AskUserQuestion tool to present multiple choice questions about colors, layouts, typography, and interaction patterns. Built a preference profile before writing any code.
Why It Works:
Inverts the typical flow. Instead of you trying to describe what you want, Claude interrogates you with specific options. Multiple choice format forces concrete decisions rather than vague preferences.
4. Handoff Points Prompt
"Create HUMAN-*.md files in root for what you need from me. Report and STOP!"
Context
Working on a complex project where Claude would inevitably need human input (API keys, design decisions, business logic). Wanted structured handoffs.
Result
Claude created HUMAN-*.md files like HUMAN-API-KEYS.md, HUMAN-DESIGN-DECISIONS.md with specific questions and clear instructions for what it needed to continue.
Why It Works:
"Report and STOP!" is crucial. Without it, Claude might try to proceed with assumptions. The HUMAN-* prefix makes these files easy to find and signals their purpose. Creates a clear async workflow for human-AI collaboration.
5. RALPH Loop Prompt
/ralph-loop 'Build a REST API for todos. Requirements: CRUD operations, input validation, tests. Output <promise>COMPLETE</promise> when done.' --completion-promise 'COMPLETE' --max-iterations 50
Context
Well-defined task with clear completion criteria. Wanted to let Claude iterate autonomously without manual intervention.
Result
Autonomous development loops. Claude iterated through implementation, testing, fixing, until all tests passed and it output the completion promise.
Warning:
RALPH loops can get expensive. The --max-iterations flag is a safety net. $50-100+ is possible for complex tasks. Use for well-defined tasks with automatic verification (like "all tests pass").
Why It Works:
The completion promise (<promise>COMPLETE</promise>) gives Claude a clear signal to emit when done. Requirements are specific and testable. Max iterations prevents runaway costs if the task proves harder than expected.
6. Folk Care Healthcare Platform
"Build HIPAA-compliant home healthcare software with EVV, family portals, scheduling, and billing."
Context
Major project: enterprise-grade healthcare platform that typically takes teams of developers 6-12 months. Compliance requirements, complex workflows.
Result
Enterprise-grade platform in 28 days. Production-ready EVV compliance, scheduling, billing, family portals. Now deployable by agencies at $20-30/month vs $750-2,250/user/month from vendors.
Why It Works:
This prompt was the starting point of many sessions. The key terms (HIPAA, EVV) signal compliance requirements. Feature list (portals, scheduling, billing) defines scope. Claude's domain knowledge filled in the implementation details.
Pro Tip:
For large projects, this initial prompt spawned dozens of follow-up sessions. Each session tackled a specific feature or subsystem. The prompt set the north star; subsequent prompts drilled into details.
Prompt Patterns That Work
Outcome-Oriented
Describe what should exist, not how to build it.
Explicit Completion
Use "STOP!", completion promises, or clear success criteria.
Domain Keywords
Terms like "HIPAA", "EVV", "REST API" trigger relevant domain knowledge.
Structured Handoffs
Use file conventions (HUMAN-*.md) for async human-AI collaboration.
Safety Limits
Always set max iterations or explicit stop conditions for autonomous loops.
Prompts That Don't Ship
"Make it good"
Too vague. Good by what criteria? For whom?
"Use React with hooks and Redux..."
Over-specifying implementation. Let Claude choose the right tool.
"Build me an app"
No constraints, no features, no completion criteria. Claude will ask follow-ups or guess poorly.
Prompts without stop conditions
For autonomous loops, always include max iterations or completion signals.
Ready to write prompts that ship?
I can train your team on vibe coding methodology or build your project directly.