A Claude skill for generating sequential, dependency-ordered PRD prompt sequences optimized for Replit Agent development.
Traditional PRDs don't work well with AI coding agents. They're monolithic, lack dependency ordering, and assume human interpretation. This skill generates phased prompt sequences that:
- Establish context before building (Plan Mode → Build Mode)
- Order tasks by dependency (database schema before queries)
- Create testable checkpoints after each phase
- Protect existing functionality with explicit constraints
# Personal use
git clone https://github.com/YOUR_USERNAME/replit-prd-skill.git ~/.claude/skills/replit-prd
# Project use (committed to repo)
git clone https://github.com/YOUR_USERNAME/replit-prd-skill.git .claude/skills/replit-prdUpload the skill folder to your Claude.ai account under Skills settings.
The skill activates automatically when you mention:
- "build with Replit"
- "Replit Agent"
- "Replit PRD"
- "vibe coding"
- "create a PRD for Replit"
"I want to build an audio library app with Replit Agent. Can you create a PRD?"
Claude will:
- Research current Replit capabilities (mandatory web search)
- Clarify your requirements and constraints
- Sequence features by dependency order
- Deliver Prompt 0 (Plan Mode) + Prompts 1-N (Build Mode)
- Adapt if you return with changes mid-project
┌─────────────┐ ┌─────────────┐ ┌─────────────┐
│ Research │ ──▶ │ Clarify │ ──▶ │ Sequence │
│ (search) │ │ (gather) │ │ (order) │
└─────────────┘ └─────────────┘ └─────────────┘
│
┌───────────────────────────────────────┘
▼
┌─────────────┐ ┌─────────────┐
│ Deliver │ ──▶ │ Adapt │
│ (prompts) │ │ (changes) │
└─────────────┘ └─────────────┘
| Principle | Description |
|---|---|
| Research before recommending | Always search current Replit capabilities |
| Native over external | Prefer Replit Database and App Storage |
| Plan Mode first | Establish context before building |
| Sequence by dependency | Foundations before features |
| Testable phases | Each prompt ends with verifiable functionality |
| Checkpoint discipline | Recommend saves after each phase |
Plan Mode: Discuss architecture without modifying code. No charges. Use for initial planning (Prompt 0).
Build Mode: Write and modify code. Creates automatic checkpoints. Use for implementation (Prompts 1-N).
replit-prd/
├── SKILL.md # Main skill instructions
├── references/
│ ├── research.md # Pre-PRD research checklist
│ ├── templates.md # Prompt structures and examples
│ └── adaptation.md # Mid-project change patterns
├── marketplace.json # SkillsMP metadata
├── README.md
└── LICENSE
The skill generates copyable prompt blocks:
## Prompt 1: Foundation
> **Use Build Mode**
[OBJECTIVE]
Set up project structure, database schema, and storage.
[REQUIREMENTS]
1. Initialize React + Vite frontend with Express backend
2. Configure Replit Database with Drizzle ORM
3. Create all database tables
[DO NOT CHANGE]
- (none for first prompt)
**Acceptance Criteria:**
- [ ] App runs without errors
- [ ] Database tables accessible
- [ ] Layout displays correctly
**Create checkpoint: "Foundation"**This skill was developed based on research into agentic coding practices (June 2025 – January 2026) and documented in the article "Product Specification as Source Code: A PM's Guide to Writing PRDs for AI Agentic Development."
Key findings:
- Sequential, phased approaches outperform monolithic PRDs
- Research-before-specification prevents integration failures
- Testable checkpoints enable confident rollback
- Explicit protection patterns prevent AI scope creep
Contributions welcome! Please:
- Fork the repository
- Create a feature branch
- Test with Claude Code or Claude.ai
- Submit a pull request
MIT License — see LICENSE for details.
Created by haberlah
Compatible with Claude Code, OpenAI Codex CLI, and other tools supporting the SKILL.md standard.