How to Prepare a PRD for Vibe Coding

Week 3 Extra Material — Deep dive into writing effective PRDs for AI-assisted development

---

Overview

This extra lesson provides comprehensive guidance on writing Product Requirement Documents (PRDs) that work well with AI coding tools like Claude Code and Gemini CLI.

The core insight: AI cannot infer from omission. What you don't explicitly specify, AI will invent. A well-written PRD prevents scope creep, hallucinated features, and wasted iterations.

---

Materials

Resource	Description
Slides (PDF)	Presentation covering all PRD sections
Checklist	One-page printable quality checklist
Examples	24 complete PRD examples

---

The 7 Sections of a PRD

1. Overview

Purpose: Define What, Who, and Why in one sentence each

| **What** | A command-line todo app for managing tasks |
| **Who** | Developers who prefer terminal workflows |
| **Why** | Quick task management without leaving the terminal |

Common mistake: Vague descriptions like "An app for users"

---

2. Core Features (MVP)

Purpose: 3-5 specific features with examples

1. **Add Task:** `todo add "Buy groceries"` → Creates task with auto ID
2. **List Tasks:** `todo list` → Shows all tasks with [ ] or [x]
3. **Complete Task:** `todo done 3` → Marks task #3 as complete

Common mistake: Vague features like "Add things" or "Make it work well"

---

3. Non-Goals (CRITICAL!)

Purpose: Explicitly list what you WON'T build

**Will NOT build:**
- User authentication or login
- Due dates or reminders
- Cloud sync or backup

**Will NOT use:**
- Database (SQLite, PostgreSQL, etc.)
- External APIs
- Rich terminal UI libraries

Why this matters: AI reads your PRD and wonders "Should I add authentication?" Without Non-Goals, it will guess (usually wrong).

Rule: If you wouldn't build it in v1, list it as a Non-Goal.

---

4. Technical Constraints

Purpose: Specify exact versions and requirements

| **Language** | Python 3.11+ |
| **Framework** | None (stdlib only) |
| **Dependencies** | None |
| **Data Storage** | JSON at ~/.todo/tasks.json |
| **Testing** | pytest, minimum 3 tests |

Common mistake: Not specifying versions or allowing "any framework"

---

5. Success Criteria

Purpose: Testable checkboxes for "done"

- [ ] `todo add "text"` creates task with unique ID
- [ ] `todo list` shows all tasks with status
- [ ] Data persists between sessions
- [ ] All tests pass

Common mistake: Vague criteria like "works well" or "is fast"

---

6. Implementation Phases

Purpose: Step-by-step build plan with verification

### Phase 1: Core Data Model
**Goal:** Set up project and implement task storage

**Tasks:**
1. Create project structure
2. Define Task dataclass
3. Implement JSON read/write

**Verification:**
python todo.py add "Test"
cat ~/.todo/tasks.json  # Shows task

**Deliverables:** todo.py, JSON utilities

Common mistake: Missing verification commands

---

7. Agent Rules

Purpose: AI decision-making guardrails

| Always | Ask First | Never |
|--------|-----------|-------|
| Validate task ID | Before schema change | Use external packages |
| Handle errors | Before new commands | Add features not in PRD |
| Show help text | Before refactoring | Skip tests |

What each column means:

• Always: Do this automatically
• Ask First: Get human approval before doing
• Never: Hard stop, don't do this

---

Common Mistakes

Mistake 1: Vague Features

❌ "Make it work well" ✅ todo list shows all tasks with [ ] or [x] prefix

Mistake 2: Missing Non-Goals

❌ (No Non-Goals section) ✅ "Will NOT build: authentication, database, cloud sync..."

Mistake 3: No Verification

❌ "Phase 1: Set up project" ✅ "Verification: python todo.py add 'Test' → Shows task ID"

---

PRD Quality Checklist

Before starting to code, verify:

• [ ] Overview has What/Who/Why (all three)
• [ ] Features are numbered (3-5 max)
• [ ] Non-Goals list at least 5 things you WON'T build
• [ ] Technical constraints specify exact versions
• [ ] Success criteria are testable
• [ ] Each phase has verification commands
• [ ] Agent Rules table has all three columns

If ANY is missing, fix it before coding.

---

Example PRDs

See templates/examples/ for 24 complete examples:

Beginner (2-3 hours):

• Pomodoro Timer, Password Generator, Flashcard Quiz

Intermediate (4-8 hours):

• Recipe API, Bookmark Manager, Habit Tracker

Advanced (1-4 weeks):

• Real-time Chat, Finance Dashboard, SaaS Starter

---

Key Takeaways

1. AI cannot infer from omission — Be explicit about everything
2. Non-Goals are critical — List what you WON'T build
3. Use Agent Rules — Define Always/Ask First/Never boundaries
4. Include verification — Every phase needs test commands
5. Be specific — Commands with expected output, not vague descriptions

---

Part of Vibe Coding 101: From Idea to Shipped Product