What is Goodhart's Law in software development?

It is the principle that when a metric (like test coverage or lines of code) becomes the target, developers find ways to hit that number without necessarily improving code quality, often leading to 'gaming' the system.

Can this skill be used for AI and machine learning optimization?

Yes, it is highly effective for red-teaming reward functions in machine learning to prevent models from finding 'shortcuts' or pathological optimizations that satisfy the mathematical target but fail the intended goal.

How does this skill help with KPI design?

It prompts Claude to identify potential exploits in your proposed metrics and suggests 'metric portfolios' where different numbers balance each other out, such as pairing Speed with Quality and Customer Satisfaction.

When should I apply Goodhart's Law patterns?

Apply these patterns whenever you are setting up new reporting, performance reviews, SLAs, or automated optimization systems where behavior is likely to adapt to the specific measurement being tracked.

Goodhart's Law Thinking Pattern

Name: Goodhart's Law Thinking Pattern
Author: lev-os

bylev-os

0•

分析と監視

Prevents metric gaming and unintended optimization by identifying when measures stop being effective proxies for goals.

This skill integrates the Goodhart's Law framework into Claude's reasoning to help developers and managers design robust measurement systems that resist manipulation. It provides structured strategies to anticipate how KPIs or performance metrics might be gamed, suggests balanced metric portfolios to create necessary tension between competing priorities, and helps identify red flags where optimization efforts are actually harming underlying objectives. It is particularly useful when establishing software quality standards, designing automated optimization loops, or setting up team performance evaluations where human or algorithmic behavior will adapt to the measurement.

主な機能

01Recommends multi-metric portfolios to balance speed, quality, and satisfaction

02Guides the evolution of metrics to prevent stale measurement systems

030 GitHub stars

04Provides implementation patterns for guardrails and counterbalancing constraints

05Analyzes system health for signs of metric-induced goal degradation

06Red-teams metrics to identify potential gaming strategies and exploits

ユースケース

01Setting up A/B testing frameworks and algorithmic reward functions in ML

02Designing software quality metrics like test coverage, bug counts, or velocity

03Creating incentive structures and KPIs for engineering and product teams

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add lev-os/agents goodharts-law

For use in Claude.ai and ChatGPT

Download Skill