Book a free call
TrainingProductsTeamSupportContactLabBlog Book a free call
Lab build · October 18, 2026

AI Skill Lab: Recruiter

One job opening. 100 applications. Most people drop them all into one big AI chat prompt and ask for the top 10. Watch how often the best candidate makes the list. Then slot in the parts of a custom-built AI Skill and see what changes.

What is a skill?

A chat prompt is a question you type every time. A skill is the way you do a job, written down once, so the AI does it the same way every time, one item at a time. Claude, ChatGPT and Microsoft 365 Copilot all load skills now. A skill is a small folder: a label that says when to use it, the steps, the hard limits, the files it reaches for only when it needs them, code for anything that has to be exact, and a running list of lessons from real mistakes.

Look inside the skill you just built

This is the actual skill used in our lab runs, built from the parts you selected above.


    
  


  

    
Build your own custom skill

    
Describe one routine pile of yours. This builds a starter skill you can paste into Claude, ChatGPT or Copilot today. It is built in your browser; nothing you type leaves this page.

    

      

        
        
        
        
        
 

      

      
Your starter skill appears here.

    

  


  

    
It works in the tools you already have

    
The skill you just built runs in the AI tool you already have, and every part has a version that works even when IT has locked things down. We recorded our runs through the model's API so we could repeat every setup and measure it. You don't need any of that.

    

    

  


  

    
Let us help you build a full set of skills

    
One skill is a start. Most teams have a dozen jobs like this one. We review how your team works, find the jobs worth a skill, custom-design each one around your workflows, and train your people to build, test and improve their own. Everything above works with zero connectors; when you're ready, we connect your skills to the systems you already use, inside the guardrails your IT and compliance people set.

    

      
A review of your workflowswhich jobs need a skill

      
Skills designed for your teambuilt around how you work

      
Hands-on trainingyour people build their own

      
Connections to your systemsinside IT's guardrails

    

    Book a free call
  










  





  
About this build
Built with Claude Opus 5.5 over the weekend of October 17, 2026. We made up a job, an operations manager at Kestrel Supply Co., and 400 applicants across four weeks, with an answer key written before any model saw them. Then we recorded 26,419 screenings: 18,377 through the Claude API on Claude Haiku 4.5, Claude Haiku 5.5, Claude Sonnet 5.5, and Claude Opus 5.5, and 8,042 on an open model, gpt-oss 20B, running on our own PC. The recording ran for just over 12 hours, from Saturday afternoon to early Sunday morning, with the API and the PC working in parallel: about 75 hours of model time and about 126 million tokens.
In every full-skill run we recorded, 15 across five models, the best candidate made the shortlist. One big chat prompt with the same 100 resumes missed her in 30 of 35 runs on Claude Haiku 5.5 and 18 of 35 on Claude Sonnet 5.5. Claude Opus 5.5, the largest model, found her in all 35 with one big prompt, at about 35 times the cost per run of Claude Haiku 5.5. On Claude Haiku 4.5, Claude Haiku 5.5 and gpt-oss 20B we recorded every combination of skill parts; on Claude Sonnet 5.5 and Claude Opus 5.5 we recorded seven steps, adding one part at a time, so any other mix on those two replays the closest step we recorded. The skill itself runs in the AI tool you already have, including tools IT has locked down. Building skills like this one is part of every training engagement we run, for companies, teams and individuals.
The job, the company, the city of Brenton and all 400 applicants are fictional. Every result the machine shows is a recorded run of a real model on those resumes, saved and replayed here. Nothing is generated while you watch, and nothing you type leaves the page. The answer key scores each applicant's real experience against the job's must-haves, ignoring the words they use. The machine never rejects anyone: a person decides every case. If you use AI in real hiring, local rules apply (New York City's Local Law 144, for example); talk to your counsel.
More builds in the Lab





  
  

    

      Sunlight shining down through deep blue water
      

        

          

            
Find light in the deep waters of AI

            
Let's talk! A short call will help you learn more about our training and consulting solutions. We can also discuss whether our existing licensed tools can help, or whether you need a custom build.

          

          Book a free call