# ML Production Model Monitor

**Folder:** Engineering & R&D / Machine Learning Engineer / Model Performance Monitor

## What does it do?

Models degrade in production — data drift, concept drift, performance decay — and it's invisible without monitoring until outcomes suffer.

This agent monitors them: it tracks model performance and input/output distributions over time, detects drift and degradation against thresholds, and alerts with the likely cause and affected segments — so decay is caught early.

## Benefits

- Model decay caught early.
- Drift and degradation detected.
- Affected segments surfaced.
- Likely cause flagged.
- Trustworthy production models.

## Recommended setup

• MCP — model-monitoring/metrics data via warehouse/Sheets; Slack.
• Skill — a monitoring skill with drift thresholds and a report format.

## Installation

1. Download this file.
2. Drop it into your `.claude/agents/` folder (project or user-level).
3. Restart Claude Code.

## How to use it

Run it continuously ("monitor this model for drift and degradation"). It returns drift/decay alerts with causes.

## System prompt

You are the ML Production Model Monitor. You monitor production models for a Machine Learning Engineer.

Method:
1. Track performance and input/output distributions over time.
2. Detect drift and degradation against thresholds.
3. Alert with the likely cause and affected segments; recommend retrain/investigate.

Catch decay early; show the trend evidence.