Promptfmt

promptfmt

A composable prompt formatting library with runtime parameter substitution and conditional logic.

Features

• Component-based architecture: Build prompts from composable components (role, goal, input, output, context, persona, tone, few-shots, guardrails, constraints, tasks, steps)
• Runtime parameter substitution: Use template strings with ${paramName} syntax
• Conditional logic: Include/exclude components based on runtime conditions
• Fluent API: Chain methods for intuitive prompt building
• Type-safe: Full TypeScript support

Installation

npm install promptfmt

Quick Start

import { PromptBuilder, createCondition } from 'promptfmt';

const prompt = new PromptBuilder()
  .role('You are a wise numerology guide')
  .goal('Generate a numerology interpretation for ${userName}')
  .input({
    name: '${userName}',
    birthday: '${birthday}',
    numberType: '${numberType}',
    numberValue: '${numberValue}'
  })
  .persona((params) => {
    if (params.age > 10) {
      return 'The Steady Anchor persona';
    }
    return 'The Clear Mirror persona';
  })
  .output('Write ${maxSentences} sentences')
  .constraints('Do not include predictions')
  .build({ 
    userName: 'John',
    birthday: '1990-01-01',
    numberType: 'lifePath',
    numberValue: 5,
    age: 34,
    maxSentences: 10
  });

Component Types

Role

Defines the AI's role/identity.

builder.role('You are a helpful assistant');

Goal

Defines the objective of the prompt.

builder.goal('Answer user questions');

Input

Defines input data/parameters. Can accept strings or objects.

// String format
builder.input('Question: ${question}');

// Object format (auto-formatted with "- key: value")
builder.input({
  name: '${userName}',
  age: '${age}',
  metadata: { key: 'value' } // Non-string values are JSON.stringify'd
});
// Results in:
// "- name: John
// - age: 25
// - meta