Odds and Ends
Things I've made when nobody asked me to. Cheatsheets, custom Claude skills, interactive learning components, side tools, and the occasional rebuild of someone else's work in my own voice. The throughline is design instinct meeting AI leverage. Take what's useful.
Audiences
Instructional designers L&D leaders rethinking their stack Builders and AI-curious knowledge workers
How to prompt Claude
better than 99% of people.
Seven moves that separate people who type at Claude from people who actually build with it. No fluff. Pin this somewhere visible.
Use Opus 4.7 with Extended Thinking.
Most people open Claude and never touch the model selector. Big mistake.
Pick Opus 4.7. Flip on Extended Thinking. The model now actually reasons before it answers.
The difference is night and day.
Stop prompting. Start writing files.
Your prompt library is collecting dust. You know it.
One great .md file (your voice rules, your examples, your guardrails) beats 50 saved prompts.
Drop it in a folder. Point Claude to it. Done.
Sounds like you. Not like AI.
Let Claude prompt you.
Add "Use the AskUserQuestion tool" to your prompt.
Claude generates a clickable form and asks you the right questions before it starts working.
Clarity before output.
Define taste by what you never say.
Don't tell Claude what you sound like. Tell it what you'd never sound like.
The "do nots" are your taste.
Taste is boundaries.
Set Global Instructions.
Settings → Cowork → Edit Global Instructions.
A persistent prompt Claude reads before every chat. Write the rules once. They run forever.
10-word prompts, full-voice output.
Build a folder system.
Three folders. That's it.
Organize your work and limit what Claude can see.
You want the damage contained.
Scope before output.
Connect your tools. It's free.
Go to Settings → Connectors. Slack, Google Drive, Notion, Gamma, 50+ others. Click "Add". Done.
Claude reads your actual tools mid-conversation. No copy-pasting. No screenshots. No "let me grab that real quick."
This is the unlock most people skip.