Forum Discussion
I turned my course QA checklist into a free Claude skill. What would you add?
I started this checklist in my master's program, and most of it was accessibility. Over the years I kept adding the things I actually catch when I review a course: activities that don't match the stated outcomes, quiz feedback that only says "incorrect, try again," screenshots of a UI that changed two releases ago.
The checklist worked great, but was manual. Recently I turned it into a Claude skill that checks 53 UX & LX criteria. You give it an exported course or a PowerPoint deck and it gives you back a checklist to validate, not a verdict. It answers what it can from the files and drafts an answer when it isn't sure. Anything it can't check, like keyboard navigation or how a screen reader sounds, it leaves open for you. Each finding gets a severity level based on how it affects the learner (blocked, inaccurate, adds effort, and so on), so you know what has to be fixed before launch and what can wait for the next revision.
It runs in the Claude desktop app or claude.ai with code execution turned on, so you don't need to install anything technical. It's been tested on a Mindsmith.ai course export on a Windows machine. If it doesn't recognize your export, it'll tell you and ask before trying, and I'd ✨love ✨ to hear how it handles Rise, Storyline or anything else.
It's free, and there's a sample report you can click through before trying it: https://chickadee368.github.io/learner-first-qa/
I'd really like to hear from people who QA courses for a living. What do you check that isn't on the list? And if you run it on one of your own exports, I'd love to know where it got things wrong!
🔗 to the overview: https://chickadee368.github.io/learner-first-qa/
🔗 to the repo: https://github.com/chickadee368/learner-first-qa
2 Replies
I'd add Axe core to the skill to do the accessibility checking part. It will check keyboard navigation for you, that you noted is manual in your current skill, amongst all other WCAG requirements... Another skill I use myself is an 'AI slop check' skill I created. It identifies any AI nonsense or made up facts that are not attributed in the content...
Chelsea - leaving open what it can't check is what makes this trustworthy. One item I'd add: does the final assessment make learners DO the task, or only recognise the right answer? A question bank that tests terms while the job needs judgement is a bigger gap than weak feedback text. A check that maps each assessment item to an observable on-the-job action would catch it. Do you flag objectives that have no matching practice activity?
Ozan Dagdeviren, Founder, AISA – The AI Fluency Assessment for Enterprises
(https://aisa.to)
