Claude Code Plugins

Community-maintained marketplace

Feedback

lovable-prompts

@jumski/dotfiles
0
0

Use when user asks to "create lovable prompts", "convert PRD to lovable", "generate lovable sequence", or provides a PRD file and asks to break it into implementation prompts. Converts PRDs into sequenced Lovable prompts with frontend mocks first, then backend integration.

Install Skill

1Download skill
2Enable skills in Claude

Open claude.ai/settings/capabilities and find the "Skills" section

3Upload to Claude

Click "Upload skill" and select the downloaded ZIP file

Note: Please verify skill by going through its instructions before using it.

SKILL.md

name lovable-prompts
description Use when user asks to "create lovable prompts", "convert PRD to lovable", "generate lovable sequence", or provides a PRD file and asks to break it into implementation prompts. Converts PRDs into sequenced Lovable prompts with frontend mocks first, then backend integration.

Lovable Prompts Generator

Convert Product Requirement Documents into a sequence of Lovable prompts that implement features step-by-step with clear, human-verifiable deliverables.

Table of Contents

Core Principles

Frontend-First with Mocks: Build UI components with dummy data/mock adapters first, then replace with real implementations in later prompts.

Incremental & Focused: Each prompt tackles ONE feature or component. Never ask for entire complex apps in one prompt.

Human-Verifiable Deliverables: Every prompt must include a checklist that both the user and Lovable can verify before proceeding.

Scope-Limited: Each prompt explicitly states what NOT to touch to prevent unintended changes.

Process

1. Read & Analyze PRD

  • Read the PRD file provided by user
  • Extract:
    • Tech stack (frameworks, libraries, backend)
    • Core features (list of capabilities)
    • In-scope (what to build)
    • Out-of-scope (what NOT to build)
    • User flows (how users interact)
    • Integrations needed (APIs, external services)

2. Identify Integrations

Map PRD requirements to Lovable integrations using [Integration Name] syntax:

  • AI/LLMs: [AI model], [OpenAI], [Anthropic]
  • UI Components: [shadcn/ui], [Chakra UI]
  • Payments: [Stripe]
  • Backend: [Supabase]
  • Maps: [Mapbox], [Google Maps]
  • Others: See prompt-integrations.md reference

Flag integrations that need API keys for setup prompts.

3. Sequence Implementation Phases

Break implementation into ordered phases:

Phase 0: Project Setup (1 prompt)

  • Create Knowledge Base entry with full PRD
  • Define tech stack and project structure
  • List all integrations needed with API key requirements
  • NO coding yet, just planning in Chat Mode

Phase 1-N: Frontend Mocks (1 prompt per feature/component)

  • Build UI components with TypeScript types
  • Use dummy data (hardcoded arrays/objects)
  • Create mock adapter classes/functions for future backend calls
  • Focus on ONE feature per prompt
  • Each includes clear deliverables checklist

Phase N+1-M: Backend Integration (1 prompt per integration)

  • Replace mocks with real implementations
  • Integrate actual APIs, databases, auth
  • Keep frontend behavior identical
  • Test and verify each integration

Phase M+1: Polish & Refinement (if needed)

  • Responsiveness fixes
  • UI/UX improvements
  • Error handling enhancements

4. Generate Numbered Prompt Files

Create files in current directory:

  • 00-setup-knowledge-base.md - Initial setup prompt
  • 01-dashboard-mock.md - First feature mock
  • 02-user-profile-mock.md - Second feature mock
  • 03-integrate-supabase-auth.md - Backend integration
  • etc.

Each file is self-contained - user can copy-paste directly into Lovable.

5. Use Structured Prompt Format

Every prompt must follow this structure (see prompt-template.md):

## Context
[Tech stack, project background, link to Knowledge Base]

## Task
[Specific deliverable for this prompt]

## Guidelines
- [Key implementation details]
- [Patterns to follow]
- [Integration syntax if needed]

## Constraints
- **Do NOT modify**: [list files/features to leave untouched]
- **Focus only on**: [specific component/page]
- **Use [Integration Name] for**: [external services]

## Deliverables Checklist
- [ ] Human-readable verification step 1
- [ ] Human-readable verification step 2
- [ ] Human-readable verification step 3

Lovable Best Practices

CLEAR Framework

  • Concise: Direct language, no fluff
  • Logical: Step-by-step, ordered instructions
  • Explicit: State exactly what to do and NOT do
  • Adaptive: Acknowledge follow-up prompts will refine
  • Reflective: Include notes on what to verify after

Chat Mode vs Default Mode

  • Phase 0 (Setup): Use Chat Mode - planning only, no code changes
  • Phases 1-N (Implementation): Use Default Mode - direct building
  • Debugging: Use Chat Mode - analyze before fixing

State the recommended mode at the top of each prompt.

Scope Limiting

ALWAYS include explicit scope constraints:

Do NOT modify the following files:
- `AuthProvider.tsx` (authentication working correctly)
- `api/payments.ts` (payment logic is stable)

Focus changes ONLY on:
- `Dashboard.tsx` and related dashboard components

Hallucination Prevention

  • Reference the Knowledge Base/PRD in every prompt
  • Provide exact field names, types, data structures
  • Include example data formats when relevant
  • Tell Lovable to ask for clarification if uncertain

Integration Syntax

Use backtick notation: [Integration Name]

Example: "Use [Supabase] for authentication and database" Example: "Generate images with [AI model]"

Prompt Generation Rules

1. ONE feature per prompt - never combine unrelated features 2. Frontend mocks MUST use dummy data or mock adapter classes 3. Every prompt MUST have explicit "Do NOT modify" constraints 4. Deliverables MUST be human-verifiable (not just "code works") 5. Reference Knowledge Base in every prompt after Phase 0 6. Use proper `[Integration Name]` syntax for all external services 7. Files numbered sequentially: 00, 01, 02... with descriptive names 8. Each prompt file is self-contained and copy-pasteable

Validation

After generating all prompts, verify:

  • Phase 0 creates Knowledge Base and uses Chat Mode
  • Frontend prompts use mocks/dummy data only
  • Backend prompts clearly state "replace mock with real implementation"
  • Each prompt has 3-5 human-verifiable deliverables
  • Every prompt includes explicit scope constraints
  • Integration syntax [Name] used consistently
  • Files numbered sequentially with clear names
  • Total prompts reasonable (typically 5-15 for most PRDs)

Resources