TOOLS & SKILLS
Claude Is About To Score Your AI Skills 🛠️
An AI Fluency Scorecard just leaked as a feature coming to Claude's settings, and it grades how good you actually are at using AI. Here's what it looks like.
Sources: TestingCatalog: the AI Fluency scorecard leak → · Anthropic: the AI Fluency Index →
Claude is about to start grading you, not your work, but on how good you actually are at using AI. The AI Fluency Scorecard was just spotted as a feature coming to Claude's settings: it scans your activity across Chat, Cowork, and Claude Code, scores you on 11 observable habits, and hands you a number out of 11 plus where to improve.
Here's roughly what it looks like:
What The Scorecard Looks Like
Illustration based on the leaked feature (via TestingCatalog) plus Anthropic's public AI Fluency framework. Exact design, habit grouping, and scoring may differ in the final release.
The 11 habits group under three skills from Anthropic's AI Fluency framework, Delegation, Description, and Discernment. Below is every habit broken down: what it means, what scores well versus poorly, and exactly how to level it up, the part the video didn't have room for.
Delegation: What You Hand Off
Delegation is everything you get right before you type, knowing the goal and handing the AI the right job with the right context.
Description: How You Frame It
Description is how precisely you frame the request. This is where vague inputs create vague outputs, and where most people lose the most points.
Discernment: How You Pressure-Test It
Discernment is quality control, the habit that separates fluent users from people who paste whatever they're handed. Anthropic's own research found people iterate plenty but question the output far too little, which makes this the rarest skill and the fastest way to stand out.
The pattern almost everyone has
Most people are decent at Description and weak at Discernment. If you fix one thing this week, start questioning the output, it's the habit with the most upside on the scorecard and in real life.
Score Yourself Before Claude Does
You don't have to wait for the feature to ship. Paste this into Claude and have it grade your real habits against the same 11, with a score and a plan to improve.
Open a recent Claude chat where you did real work, then paste this:
The One Prompt That Aces All Three
Once you know the 11 habits, you don't have to think about them every time. This skeleton bakes delegation, description, and discernment into one reusable prompt, fill it in for anything that matters:
Save this and reuse it for any real task:
The real unlock of AI agents is not full autonomy, but human-in-the-loop systems design. Structure your agent inputs with explicit constraints, negative prompts, and automated test checkpoints. When building tools, keep token consumption lean by caching system prompts and isolating tasks into specialized sub-agents.