Creating Custom Prompts

This guide walks you through creating your own custom prompts in Nexus.

Prompt Structure

A prompt consists of several components:

interface Prompt {
  name: string; // Display name
  slug: string; // URL-friendly identifier
  description: string; // What this prompt does
  systemPrompt: string; // AI instructions
  parentPromptId?: string; // Optional parent to extend
  libraries: string[]; // Documentation to include
  skills: string[]; // Skills to enable
  mcpServers: string[]; // MCP servers to suggest
  preferences: {
    verbosity: "concise" | "balanced" | "detailed";
    codeStyle: "minimal" | "documented" | "verbose";
    responseFormat: "full" | "compact" | "code-only" | "summary";
    useEmojis: boolean;
    preferredLanguage?: string;
  };
  category: string; // For organization
  tags: string[]; // Searchable tags
}

Creating via Dashboard

1. Navigate to Prompts Dashboard
2. Click "Create Prompt"
3. Fill in the form:
   • Name: Human-readable name
   • Slug: URL-safe identifier (auto-generated from name)
   • Description: What the prompt is for
   • System Prompt: Instructions for the AI
   • Parent Prompt: Optional prompt to extend
4. Configure libraries, skills, and preferences
5. Click "Create"

Creating via CLI

Interactive Mode

nexus prompt create --interactive

This guides you through each option with prompts.

Direct Creation

nexus prompt create \
  --name "React Testing Expert" \
  --slug "react-testing" \
  --description "Focused on React component and integration testing" \
  --parent testing-qa-engineer \
  --libraries react,vitest,testing-library \
  --prompt "You are an expert in React testing..."

Writing Effective System Prompts

Structure

A good system prompt includes:

1. Role Definition: Who the AI is
2. Expertise Areas: What it knows well
3. Behavioral Guidelines: How it should respond
4. Constraints: What to avoid
5. Examples: Sample interactions (optional)

Example System Prompt

You are an expert React and TypeScript developer with deep knowledge of:

- React 18+ features (hooks, suspense, concurrent rendering)
- TypeScript 5+ with strict type checking
- Modern state management (Zustand, Jotai, React Query)
- Testing with Vitest and React Testing Library

## Guidelines

- Prefer functional components with hooks over class components
- Always use TypeScript with explicit return types for exported functions
- Suggest appropriate error boundaries for production code
- Include accessibility attributes (aria-\*, role) in UI components

## Code Style

- Use named exports over default exports
- Destructure props in function parameters
- Prefer early returns for guard clauses
- Add JSDoc comments for public APIs

## Constraints

- Do not suggest deprecated patterns (componentWillMount, etc.)
- Avoid inline styles; prefer CSS modules or Tailwind
- Do not use `any` type; prefer `unknown` with type guards

Extending Existing Prompts

Using Inheritance

Extend a prompt to add or override configuration:

nexus prompt create \
  --name "React E2E Testing" \
  --parent react-testing \
  --libraries playwright \
  --prompt "Additional focus on end-to-end testing with Playwright..."

Inheritance Rules

1. System Prompts: Child prompt appended to parent
2. Libraries: Merged (union of both)
3. Skills: Merged (union of both)
4. Preferences: Child overrides parent
5. Max Depth: 3 levels of inheritance

Example Inheritance Chain

base-developer (Level 0)
  └── typescript-expert (Level 1)
       └── react-typescript (Level 2)
            └── react-testing (Level 3) ← Maximum depth

Configuring Preferences

Verbosity

Level	Description
concise	Minimal explanations, focus on code
balanced	Moderate explanations with code
detailed	Thorough explanations and context

Code Style

Style	Description
minimal	Just the essential code
documented	Code with inline comments
verbose	Extensive comments and examples

Response Format

Format	Description
full	Complete response with metadata
compact	Essential information only
code-only	Just code blocks
summary	Brief overview

Including Libraries

Add documentation libraries to your prompt:

# Find available libraries
nexus docs list

# Add to prompt
nexus prompt create \
  --libraries react,typescript,tailwindcss,prisma

Libraries provide pre-indexed documentation that the AI can query.

Including Skills

Add specialized skills:

# Find available skills
nexus skills list

# Add to prompt
nexus prompt create \
  --skills code-review,security-audit,performance-optimization

Best Practices

1. Start Specific

Create focused prompts rather than trying to cover everything:

Good: "React Form Validation Expert"
  └── Focused on forms, validation, error handling

Bad: "Full-Stack Everything Expert"
  └── Too broad to be useful

2. Use Inheritance Wisely

Build a hierarchy of prompts:

frontend-base
  ├── react-expert
  │   ├── react-forms
  │   └── react-data-fetching
  └── vue-expert
      └── vue-composition

3. Test Your Prompts

After creating a prompt:

1. Activate it: nexus prompt activate my-prompt
2. Try various queries
3. Verify the AI behavior matches expectations
4. Refine the system prompt as needed

4. Document Your Prompts

Add clear descriptions so others (or future you) understand:

nexus prompt create \
  --description "For greenfield React projects using TypeScript 5+,
                 React Query for data, and Tailwind for styling.
                 Assumes Vite as build tool."

Sharing Prompts

You can make prompts public for others to use:

1. Set isPublic: true when creating
2. Others can search and fork your prompt
3. Forked prompts can be customized

To share a prompt configuration:

1. Download the prompt: nexus prompt download my-prompt
2. Share the PROMPT.md file
3. Others can import or recreate the prompt

Troubleshooting

Prompt Not Activating

Check that you're authenticated:

nexus auth status

Inheritance Not Working

Verify parent prompt exists:

nexus prompt get parent-prompt-slug

System Prompt Too Long

Keep prompts focused. If you need extensive instructions, consider:

1. Using inheritance to split into layers
2. Linking to external documentation
3. Using skills for specialized knowledge