20 Secret Codes
for Claude

The escalation ladder. Type think, think hard, think harder, or ultrathink and each one gives Claude a bigger thinking budget before it responds. ultrathink is the max. Use it on hard problems where the first answer usually is not good enough.

Type this once in Claude Code and it stays in deep-thinking mode for the whole session, not just one message. Set it to max before a complex build and every response gets the full reasoning budget.

The anti-hallucination line. Add "if you’re not sure, just say I don’t know" to a prompt, or drop it into your global instructions so it’s always on, and Claude stops confidently making things up.

Stops Claude from guessing what you want. Say this before a big task and instead of a generic answer, it interviews you one question at a time until it actually understands the job.

The app version of ultrathink. Flip the thinking toggle near the message box and Claude reasons through the problem before answering, and you can watch it think. Leave it on for anything analytical.

Stop getting walls of text. Ask Claude to build something "as an artifact" and instead of describing it, it builds a live, working version in a panel next to the chat: a calculator, a tool, a one-page app, a doc you can edit.

Your own custom Claude. Create a Project, upload your docs as its knowledge, and add instructions. Now every chat inside it already knows your context. No re-explaining yourself every time.

Make Claude write like you instead of like a robot. Pick a built-in style (Concise, Explanatory, Formal) or create a custom one trained on your own writing. Set it once and every response matches your voice.

For the big questions. Turn on Research and Claude goes off and reads dozens of sources, then comes back with a real, cited answer instead of a quick guess. Use it when being right actually matters.

Plug your actual tools into Claude. Connectors let Claude reach into Gmail, Google Drive, your calendar, Notion, and hundreds of other apps directly. It’s not guessing, it’s working with your real stuff.

The reset button. Wipes the conversation so Claude starts fresh, but keeps every file change it already made. When a session goes off the rails, /clear and re-ask cleanly instead of fighting the context.

The undo button for AI. Tap Escape twice and you get a rewind menu. Jump back to any earlier point and restore the conversation, your code, or both. Claude broke something? Rewind instead of starting over.

Type @ and start the filename, and Claude instantly loads that exact file into the conversation. No copy-pasting, no "here’s my code." Just @filename and it’s in context.

Plan Mode. Make Claude plan before it touches anything. Hit Shift+Tab to cycle into Plan Mode and Claude maps out exactly what it’s going to do, then waits for your okay before changing a single file. Essential for big or risky tasks.

Stop typing. Tap the mic and just talk. It’s faster than your thumbs, and you explain messy, complicated stuff way better out loud than in a tidy little text box.

The one-time setup almost nobody does. In settings there’s a box for what Claude should always know about you and how to respond. Fill it in once and every chat starts already knowing your job, your style, your preferences.

Same idea as Custom instructions, but for Cowork and Code. A file Claude reads at the start of every session: your standards, your context, how you like things done. Write it once and stop re-explaining yourself. Type /init and Claude drafts the first version by reading your project.

Save a memory mid-task. Start a line with # and Claude saves whatever you type straight into your CLAUDE.md. No stopping to edit files. Teach it something once and it remembers from then on.

Turn one Claude into a whole team. Add "use subagents" to a big task and Claude spins up multiple agents that work in parallel. Hours of work compressed into minutes. Run /agents to build your own specialized ones.

The forever button. After Claude nails a task, have it save the whole workflow as a Skill: a reusable command you can run anytime with one word. Build it once, run it forever.