prompt-template-builder 🚀

Ultra-fast, zero-dependency, type-safe prompt template engine for AI agents and LLMs.
Designed for Anthropic Claude, OpenAI GPT-4, Gemini, Ollama, and Vercel AI SDK integrations.

---

Features

• ⚡ Zero External Dependencies: Lightweight footprint, compatible with Node.js, Bun, Deno, and Edge Runtimes.
• 🎯 Type-Safe Builder Pattern: Fluent API to assemble multi-turn conversation messages (system, user, assistant, developer).
• 🛡️ Prompt Injection Guardrails: Built-in XML content escaping and template variable sanitization.
• 🏷️ First-Class XML Tagging: Seamlessly insert Anthropic/OpenAI recommended <context>, <instructions>, and <documents> tags.
• 🔢 Token Estimator: Quick heuristic token estimation without heavy tokenizer bundles.

---

Installation

npm install prompt-template-builder
# or
pnpm add prompt-template-builder
# or
yarn add prompt-template-builder

---

Quick Start

import { createPromptBuilder } from 'prompt-template-builder';

const prompt = createPromptBuilder({ escapeXml: true })
  .system('You are an expert {{language}} developer assisting with code review.', {
    language: 'TypeScript',
  })
  .user('Please review the following module:')
  .xmlSection('code', 'function add(a: number, b: number): number { return a + b; }', {
    filename: 'math.ts',
  })
  .buildMessages();

console.log(prompt);
/*
Output:
[
  { role: 'system', content: 'You are an expert TypeScript developer assisting with code review.' },
  { role: 'user', content: 'Please review the following module:\n\n<code filename="math.ts">function add(a: number, b: number): number { return a + b; }</code>' }
]
*/

---

Advanced Usage

Variable Interpolation Styles

You can specify custom variable delimiters:

// {{var}} (default), ${var}, or [var]
const builder = createPromptBuilder({ delimiterStyle: 'dollar-brace' });

builder.user('Hello ${name}, welcome to ${service}!', {
  name: 'Swaraj',
  service: 'Open Source',
});

Integration with OpenAI & Anthropic SDKs

import Anthropic from '@anthropic-ai/sdk';
import { createPromptBuilder } from 'prompt-template-builder';

const builder = createPromptBuilder()
  .system('Synthesize the user query concisely.')
  .user('Explain quantum computing in 2 sentences.');

const client = new Anthropic();
const response = await client.messages.create({
  model: 'claude-3-5-sonnet-20241022',
  max_tokens: 300,
  messages: builder.buildMessages().filter(m => m.role !== 'system') as any,
  system: builder.buildMessages().find(m => m.role === 'system')?.content,
});

---

API Reference

createPromptBuilder(options?: PromptTemplateOptions)

Options:

• delimiterStyle: 'double-brace' | 'dollar-brace' | 'square-bracket'
• stripUnhandledVariables: boolean (Default: false)
• escapeXml: boolean (Default: false)

Methods

• .system(template, variables?): Adds a system message.
• .developer(template, variables?): Adds a developer message (for OpenAI o-series models).
• .user(template, variables?): Adds a user message.
• .assistant(template, variables?): Adds an assistant response.
• .xmlSection(tagName, content, attributes?): Appends structured XML tag block.
• .estimateTokens(): Returns estimated token count.
• .buildMessages(): Returns standard message array.
• .buildText(): Returns single joined text prompt.
• .clear(): Clears history.

---

Contributing

Contributions are warmly welcome! Please see CONTRIBUTING.md for details.

---

License

MIT © Swaraj Jakanoor