What You’ll Learn

Claude Skills have always been a powerful way to give your AI assistant a set of instructions — like handing an employee a job description. But there was always a problem: you built the skill, crossed your fingers, and hoped it worked. There was no test. No interview. No way to know if the skill actually improved your outputs before you committed to using it.

Skills 2.0 in Claude Co-work changes that completely. This tutorial walks you through how the new evaluation system works, what the “hiring manager” concept means in practice, and how to use it to build skills that actually do what you need them to do.

• Understand the core difference between Skills 1.0 and Skills 2.0
• See how the 6-agent parallel testing system works
• Learn how to define pass/fail criteria before building any skill
• Know what you need to get started with Skills 2.0

The Problem with Skills 1.0: Building in the Dark

When you created a skill the old way — through the web version of Claude — the process was simple but risky. You’d write a set of instructions, save the skill, and then try it out on a real task. If it worked, great. If it didn’t, you had no idea why.

Think of it like hiring an employee without doing an interview. You write a job description, hand it to someone, and on day one you find out whether they can actually do the job. That’s a lot of wasted time if they can’t.

“It’s like you build the skill, you hire the employee, and hope to God it does exactly what you want it to do.” — James, 02:15

The concept of skills was right. The quality control was missing. Skills 2.0 solves that.

? In Plain English: A “skill” is a saved set of instructions you give to Claude so it always behaves a specific way for a specific task — like a recipe it follows every time. Skills 1.0 meant writing the recipe and hoping it tasted good without ever testing it first.

How Skills 2.0 Works: The Hiring Manager Model

Skills 2.0 introduces an evaluation loop before your skill ever gets deployed. Instead of building and hoping, you now have a hiring manager built into the process — one that pre-screens, tests, and scores your skill against real tasks before you start using it.

Here’s how the process works step-by-step:

Step 1: Claude Asks Clarifying Questions

Before building anything, the skill creator asks you what the skill needs to do and — critically — what “good” looks like. You define what a pass is. You define what a fail is. This forces you to think through the outcome before writing a single instruction.

✓ Check Your Work: Can you describe what a perfect output from this skill looks like? If you can’t, you’re not ready to build yet.

Step 2: Six Sub-Agents Run in Parallel

Once you’ve defined the criteria, the system spins up six sub-agents simultaneously — three using your new skill and three running without it. Both groups get the same test prompts. Both produce outputs. The system captures all six results.

“Six tests are running in parallel right now — three with and three without. Something I would have had to do manually in Skills 1.0.” — James, 08:10

This is the key upgrade. In Skills 1.0, if you wanted to know whether your skill was making things better, you’d have to test it manually — run the prompt with the skill, run it without, compare them yourself. Now the system does all of that automatically in the time it used to take to run one test.

Step 3: Quantitative Assertions Run Against Every Output

The system doesn’t just produce outputs — it scores them. It runs your pass/fail criteria as actual checks against every result. Did the output meet the length requirement? Did it include the required sections? Did it match the tone? Each assertion passes or fails with a clear result.

Step 4: A Browser-Based Review Tool Generates Automatically

After testing, Skills 2.0 generates a side-by-side HTML comparison tool you can open in any browser. Left column: outputs without the skill. Right column: outputs with the skill. You can see exactly what your skill changed — or didn’t change — before you decide whether to keep it.

“We now have a hiring manager that goes out, pre-sorts a whole bunch of people, gives them tests, and lets us either pass, fail, good, bad, ugly before the skill gets generated.” — James, 05:45

Step 5: Brand Guidelines and Business Context Are Baked In

Skills 2.0 also integrates your brand guidelines, your ICP (ideal customer profile), and your business context directly into the evaluation. Your skills aren’t just tested against generic quality — they’re tested against your specific standards. The skill knows who you’re talking to and what your outputs should sound like.

The Mindset Shift: Define Pass/Fail Before You Build

This is the part most people skip — and it’s the most important part of the whole system.

Before you start building any skill, you need to answer two questions:

1. What is this skill supposed to do? Not in vague terms — specifically. “Write a social media post” is not a job. “Write a LinkedIn post under 200 words that leads with a question and ends with a single CTA” is a job.
2. What does a pass look like? Define it numerically where you can. “Includes at least 3 steps,” “uses the client’s name,” “under 150 words.” The more specific your pass/fail criteria, the more useful the evaluation.

“Creating a skill for the sake of creating a skill is not the way to get an employee that is going to help your business save you time or earn money.” — James, 11:00

⚠️ Important: If you can’t define what a passing output looks like before you build, you’re not ready to build. Start with the outcome and work backwards.

What You Need to Use Skills 2.0

There are two requirements to access the Skills 2.0 evaluation system:

Requirement 1: Claude Co-work (Desktop App)

Skills 2.0 is a desktop-only feature available in Claude Co-work. It is not available in the web version of Claude. If you’re currently building skills at claude.ai, you’re still using the Skills 1.0 approach — there’s no evaluation loop, no parallel testing, no comparison tool.

You’ll need to download and install Claude Co-work to access these features.

Requirement 2: Clear Pass/Fail Criteria

The technical requirements are simple — the conceptual requirement is harder. Before you open the skill creator, you need to know what the skill is supposed to do and what success looks like. The system will ask you. Have an answer ready.

? In Plain English: Think of it like writing a rubric before grading a test. You need to know what an “A” looks like before you can grade anything.

Quick Reference: Skills 1.0 vs Skills 2.0

Your Next Steps

Here’s how to put this into practice right now:

1. Download Claude Co-work if you haven’t already. Skills 2.0 requires the desktop app.
2. Pick one skill you already use (or want to build) and write down what a passing output looks like in specific terms.
3. Open the skill creator in Co-work, answer the clarifying questions, and let the evaluation loop run.
4. Review the side-by-side comparison in the browser-based review tool before you decide to deploy.

Want 250 pre-built skills to start with? Visit trainingsites.io and join the community. Everything is ready to use with Claude Co-work the moment you arrive.