🪄 magic-need

Capture tool/data needs from AI agents — let your agent spec your product for you. Give your AI agent a tool that "does nothing" but lets it express what it's missing. The agent becomes a proxy for your best engineer and specs your integration roadmap for you.

For OpenClaw Users

This CLI is included in the magic-need skill. Install via:

clawhub install magic-need

Then use: node ~/.openclaw/skills/magic-need/scripts/cli.js "description"

Skill repo: https://github.com/guim4dev/skill-magic-need

For NPM Users

npm install -g magic-need

Or with npx (no install):

npx magic-need "API for recent deploys"

Usage

As an AI Agent

When you're executing a task and realize you need data or a tool you don't have:

magic-need "API of recent deploys for service X in the last 2 hours"

Or via code:

const { execSync } = require('child_process');
execSync('magic-need "CPU metrics for upstream auth-service"');

As a Human

# Register a need
magic-need "Need for Stripe webhook handling"

# List all needs
magic-need list

# Generate report (for cronjobs)
magic-need report

# Archive pending needs
magic-need clear

Auto-categorization

Needs are automatically categorized based on description:

| Category | Triggers | Emoji | |----------|----------|-------| | integration | api, endpoint | 🔌 | | observability | metric, log, monitor | 📊 | | devops | deploy, pipeline, ci | 🚀 | | auth | user, auth, login | 🔐 | | database | database, db, query | 🗄️ | | storage | file, storage, upload | 📁 | | general | (default) | 📝 |

Cronjob Example

Daily report at 10 PM:

# Add to crontab
crontab -e

# Add this line:
0 22 * * * /usr/local/bin/magic-need report | webhook-send https://discord.com/api/webhooks/...

Or with a Discord bot:

const report = require('child_process').execSync('magic-need report').toString();
if (!report.includes('NO_REPORT')) {
  // send to Discord channel
}

Data Storage

Needs are stored in ~/.magic-need/needs.json (global install) or ./data/needs.json (local dev).

Structure:

[
  {
    "id": "j8ldlr",
    "description": "API de deploys recentes...",
    "createdAt": "2026-03-07T18:09:18.123Z",
    "status": "pending",
    "category": "integration"
  }
]

Why?

"The insight: your agent is a great proxy for your best on-call engineer. Give it a blank canvas to express what it's missing, and it'll spec your product for you." — Sonarly

Instead of guessing which integrations to build, let your AI agent tell you what it actually needs.

License

MIT © Thiago Guimarães