@mounaji_npm/human-in-loop

Extensible human-in-the-loop toolkit for agent workflows.

It provides:

• Core prompt schema and UI inference engine (framework-agnostic)
• Prompt type registry for future extensions
• React components and hooks for rendering and collecting decisions

Install

npm install @mounaji_npm/human-in-loop

Quick start (core)

import { inferPromptUI, createPromptRegistry } from '@mounaji_npm/human-in-loop';

const prompt = {
  id: 'color-question',
  type: 'color',
  label: 'Choose primary color',
  suggestions: ['#2563eb', '#10b981', '#f59e0b'],
};

const ui = inferPromptUI(prompt);
console.log(ui);

Quick start (React)

import { HumanInLoopPanel } from '@mounaji_npm/human-in-loop';

<HumanInLoopPanel
  prompts={prompts}
  onSubmit={async (decision) => {
    console.log(decision);
  }}
/>

Supported prompt types

• single_select
• multi_select
• redacted_agent_text
• color
• text
• number
• boolean_confirm
• date
• json

Redacted agent text options

The redacted_agent_text prompt type supports the following options by default:

1. remake code from scratch
2. just fix x part
3. both of them
4. custom answer

It always renders a custom text input, independent of the selected option.

const prompt = {
  id: 'agent-redaction-response',
  type: 'redacted_agent_text',
  label: 'How should the agent proceed?',
  required: true,
  metadata: {
    customTextLabel: 'Extra instruction',
  },
};

Extending with custom types

Use createPromptRegistry() and register custom inference/renderers progressively.