Teach by showing, and explain what is on the screen

Two ways to work with what you see: show Pinchy a task once and he turns it into a skill, or drag a box around anything and he explains it.

What to ask for

“Let me show you how I file my invoices.” A small bar appears; you do the task once, add notes if you like, and press Finish.
“Watch how I do this on the website.” The same, in his browser.
“Do the invoice thing again.” He runs the skill you taught him.
Press the explain shortcut and drag a box around an error message, a chart or a figure in a PDF: he explains it in the chat.

Teaching: he notes which window is in front, the names of the buttons and fields you use, and what you write in the Note box. He never records your keystrokes. A few screenshots of the window help him understand clicks he cannot name; they are deleted as soon as the skill is written, and never taken of password managers or while Private is on.
When you press Finish he writes a skill draft. You read it, then save or discard it. Saved skills load by themselves when a later task matches.
You can teach in three places: your own desktop (say “watch what I do”, or put Teach me into his right-click menu with Customize), his browser, and his own computer (the Watch me button in each).
Explain: only the box you drag is sent to him, with an optional standing question such as “what does this error mean”.

Settings → Computer → Teach by showing: teach in my own apps, look at the app I am showing, read button and field names, let Claude tidy the steps, the model for writing skills, and “Watch me now”.
Settings → Computer → Explain the screen: switch it on (off by default), the shortcut, what to ask about the region.
Settings → Add-ons → Skills: every skill he has, switch one off or remove it.

Reading button and field names uses the desktop accessibility service (Linux). Password fields are never read.
A taught skill repeats the steps you showed. If the app changes a lot, teach it again.