Book 2 · Chapter 4: Working With Your Actual Files
This page gives you a card for deciding which files are safe to share with AI, plus starter folder structures.
This page assumes you've read the chapter. It hands you the tools, not the lessons.
Download · PDF
File-Safety Decision Card
Ask yourself
- 1Does this file contain secrets I wouldn't want Claude to see? If yes, don't connect the folder.
- 2Does this file contain data I wouldn't want in Claude's conversation logs? If yes, maybe don't ask Claude to read it, or be specific about what to read instead.
- 3Is there a business reason to have Claude read this file? If no, why are you connecting the folder?
- 4Can I read the file back after Claude's done with it, and am I comfortable with what I'm seeing? If no, that's a sign not to do it.
Claude can read
- Text files, Word docs, native PDFs (just don't include passwords in the text)
- Spreadsheets, Excel, CSV (just don't include credit card numbers in a column)
- Images and screenshots (Claude can read text in images, but shouldn't be your password manager)
- Code and technical specs
Claude shouldn't read
- Password files, API keys, tokens
- Banking passwords, login credentials
- Unencrypted financial account numbers
- Health or medical records you're trying to keep private
From the Decision Tree: Is This File Safe to Share? section.
Folder-structure starter examples
Connect only the folders where the benefit of AI access is worth the trust you place in it. Two example setups, following that rule:
Work
Connect: Client Work/ active client projects Current Invoices/ monthly anomaly checks Project [Name]/ one scoped project folder Don't connect: Personal Finance/ banking and statements stay off-limits
Personal
Connect: Family Documents/ school records, forms, kids' paperwork Home Expenses/ the household budget for analysis Don't connect: Banking/ no reason for AI to read it