43.2 Shared prompt libraries and templates

The Problem

Without a shared library, every developer reinvents common prompts:

• 5 devs = 5 different SQL generation prompts
• No consistency in error handling
• Best practices aren't shared
• Bugs fixed in one prompt aren't fixed in copies

Library Structure

prompts/
├── core/
│   ├── safety-rules.md          # Included in ALL prompts
│   ├── output-formats.md        # JSON, markdown templates
│   └── error-handling.md        # Standard error responses
├── domains/
│   ├── sql-generation.md        # SQL from natural language
│   ├── code-review.md           # Review code changes
│   ├── summarization.md         # Document summarization
│   └── classification.md        # Text classification
├── components/
│   ├── chain-of-thought.md      # Reasoning blocks
│   ├── few-shot-header.md       # Example injection
│   └── self-check.md            # Self-verification
└── config/
    ├── models.yaml              # Model configs per use case
    └── eval-sets/               # Test data per prompt

Template System

// prompt-template.ts
export function buildPrompt(templateName: string, variables: Record): string {
  const template = loadTemplate(templateName);
  const coreRules = loadTemplate('core/safety-rules');
  
  let prompt = `${coreRules}\n\n${template}`;
  
  // Variable substitution
  for (const [key, value] of Object.entries(variables)) {
    prompt = prompt.replace(new RegExp(`{{${key}}}`, 'g'), String(value));
  }
  
  return prompt;
}

// prompts/domains/sql-generation.md
`{{core/safety-rules}}

You generate SQL queries from natural language.

## Database Schema
{{schema}}

## Rules
- Only SELECT queries (no INSERT, UPDATE, DELETE)
- Use table aliases for clarity
- Always include LIMIT unless aggregating

## User Question
{{question}}

## Output
Return valid SQL only, no explanation.`

Usage

// Using the library
import { buildPrompt } from '@company/prompt-library';

const prompt = buildPrompt('domains/sql-generation', {
  schema: databaseSchema,
  question: userQuery
});

const response = await model.generateContent(prompt);

// Benefits:
// ✅ Safety rules automatically included
// ✅ Consistent format across team
// ✅ One place to update when improving
// ✅ Easy to test with standard eval set

Where to go next

• 43.3 Spec → prompt → eval → ship pipeline