Shared Claude Skills That Stop Workslop (Built From Five Answers)

The strongest predictor of AI workslop is a company pushing AI use without showing anyone what good looks like. In the Stanford and BetterUp survey of 962 desk workers, each piece cost the person who received it about 1.7 hours.

A skill writes the standard down once. It holds your best examples, rules that pass or fail, a checklist Claude runs on its own draft, one named owner and one named reviewer. Everyone on the team who runs it gets the same result.

This folder builds one in about 10 minutes from a prompt your team already runs every week. Every fix you make by hand becomes a rule, and the rework count gives you the number to take to your boss in two weeks.

What’s in the download:

• A finished launch post skill, with Attio’s real posts as the examples, so you can see what good looks like before you build your own

• make-skill, which turns any prompt your team reuses into a shared skill from five answers

• log-fix, which counts each hand fix and adds it to the skill’s checklist

• rework-count.md, to log every draft you fix and how long it took

• owners.md, listing every skill with its owner and its reviewer

• BUILD_STEPS.md with every prompt in order, and PREREQUISITES.md

• A real changelog entry to run the finished skill on, and our own outputs to compare against

Built by Will Leatherman and the team at Catalyst, who’ve built content and AI-search systems for 150+ B2B companies.