Seeing Through the Matrix

Power user moves that make you feel like you have a second nervous system wired into your computer.

Step 1: Search your entire computer

Claude can't read your mind, but it can read your files. When you can't find something, don't dig through folders. Ask Claude to grep for it.

I wrote something about pricing strategy a few weeks ago but I
can't remember where I saved it. Search my entire strategy folder
for anything mentioning "pricing" or "rate card" and show me
what you find.

Claude uses grep under the hood. It searches file contents, not just file names. It'll find that pricing note buried three folders deep in a file you forgot existed.

You can get more specific:

Search my whole home directory for any file containing
"client lunch meeting" — I know I wrote notes somewhere.

Or search by file type:

Find all markdown files in my strategy folder that were
modified in the last 7 days. Show me the file names and
the first few lines of each.

Step 2: Read anything, anywhere

Claude can read files outside your current folder. PDFs, images, screenshots, spreadsheets. If it's on your machine, Claude can look at it.

Read the PDF at ~/Downloads/proposal-v3.pdf and summarize
the key terms. What am I agreeing to?

Or point it at a screenshot:

Look at this screenshot: ~/Desktop/Screen Shot 2026-04-05.png
What's wrong with this layout? What would you change?

Or pull from your Downloads folder:

Read the CSV at ~/Downloads/campaign-metrics.csv and tell me
which campaigns are underperforming. Sort by ROI.

Step 3: Front-load context to avoid wrong turns

The biggest friction point with Claude is when it tries the wrong approach first. The fix is simple: tell Claude what you suspect before it starts exploring.

# Instead of this:
"The app is broken after the last update."

# Do this:
"The app crashes on launch after the last update. I suspect
it's the new auth flow because that's what changed. Check
the auth middleware first. Here's the error log: [paste].
Don't explore broadly until that's ruled out."

The pattern: what's broken + what you suspect + where to look first + what not to touch. Four pieces of context that save 20 minutes of wrong approaches.

Step 4: Drop images and screenshots into prompts

You think visually. So does Claude. Take a screenshot of what you want, drop it in.

Here's a screenshot of the current dashboard: [drag image in]
Here's what I want it to look like: [drag mockup in]
What needs to change? List the specific differences.

Works with mockups, wireframes, competitor screenshots, error screens, even whiteboard photos. Claude sees the image and responds to what's in it, not what you describe.

Step 5: Chain operations into pipelines

The real power is chaining multiple steps together. One prompt, multiple operations, final output.

Voice memo to structured notes

Transcribe the audio file at ~/Desktop/meeting-notes.m4a,
then extract all action items, group them by person,
and save the result to brain/00-inbox/meeting-2026-04-06.md

Research to artifact

Search my brain/10-knowledge/ folder for everything related
to pricing. Then search the web for current freelance creative
director day rates. Combine both into a one-page rate card
and save it as a clean HTML file I can print.

Competitor analysis

Go to these three websites and take a screenshot of each
homepage. Then compare them: what's each one doing well?
What's missing? How does ours compare? Save the analysis
to projects/[name]/competitive-audit.md

Step 6: Hooks

Hooks run automatically when Claude does certain things. You set them up once and forget about them.

For example: auto-format every file Claude edits. Auto-check for errors after every change. Auto-log what Claude did to a file.

# Ask Claude to set up a hook for you:
Set up a hook that automatically formats any file you edit
using prettier. Add it to my Claude Code settings.

You don't need to know the settings.json syntax. Just describe what you want to happen automatically, and Claude configures it.

Step 7: The on-call map

When you open Claude in a new project or unfamiliar codebase, don't start working immediately. Ask for a map first.

Before we do anything, give me an on-call map of this project:
- What's the tech stack?
- What's the folder structure?
- What are the last 5 git commits?
- Where are the entry points?
- What looks fragile or messy?

Use sub-agents to explore in parallel. Be fast.

This takes 30 seconds and saves you from the most common failure: Claude changing something without understanding the full picture first. Orient, then operate.

Step 8: Scope-lock your sessions

The biggest mistake is letting a session sprawl. One bug fix turns into a refactor turns into a new feature turns into "what happened to the last two hours?"

This session is ONLY for fixing the login bug. Don't refactor
anything. Don't improve adjacent code. Don't add features.
Fix the bug, verify it works, commit. That's it.

Step 9: Run your own usage insights

Claude Code tracks how you use it. You can run an analysis of your own sessions to see what's working, what's causing friction, and what to try next.

/insights

This generates a full report: which projects you work on most, your interaction style, what goes well, where things break down, and specific suggestions for your CLAUDE.md. It's like having a coach review your tape.

Step 10: The interrupt

This is the most underrated power user move. If Claude is going down the wrong path, generating code you don't want, or taking too long, hit Escape.

Then restart with tighter constraints. Don't let a bad approach run for 2 minutes when you knew it was wrong at second 10.

The power move isn't knowing every command. It's knowing when to stop, redirect, and try again with better context. That's judgment. And that's what you already have.