Cindy Zhu.
← all free guides
Power moves Claude Code

4 Claude Code prompts that turn it into a super-employee

Hey, it's Cindy 🌱 These are the 4 prompts that stop you using Claude like a chatbot and start using it like a teammate that runs parts of your week: a memory, a decision log, an inbox helper, and a task runner. Copy them straight in. I checked every mechanism against Anthropic's docs on 2026-07-14 and kept it honest about what's on-demand vs scheduled.

One honest frame before the prompts: Claude works inside a session you start (and, if you want, on a schedule you set up). It is a copilot you drive, not a bot that silently runs your business. That's the accurate, non-hype version, and it's still a huge upgrade.

copy these

Prompt 1: give it a memory 🧠

Claude forgets between chats. Fix it once. CLAUDE.md is auto-loaded at the start of every session, so point it at your own notes.

🧠 the memory system 
Create a persistent memory system for yourself in a /memory folder with files:
decisions.md, people.md, preferences.md, user.md. Then write a CLAUDE.md that reads
those files at the start of every session and updates them as we work.
Honest note: CLAUDE.md is strong context, not an enforced rule, and Claude updates the files as you work (not a guaranteed "auto-save at session end"). Docs: code.claude.com/docs/en/memory.
keep yourself honest

Prompt 2: make it log your decisions 📓

Build your own track record so you stop repeating mistakes.

📓 the decision log 
Create a decision log. Whenever I describe a decision, append a row to decisions.csv
with: date, decision, reasoning, expected outcome, and a 30-day review date. Also give
me a review command that surfaces only the entries whose review date has passed.

To get the reviews surfaced automatically, set a routine (cloud-scheduled, laptop closed, Pro/Max and up, currently in preview) or just run the review command yourself.

the big time-saver

Prompt 3: let it triage your inbox 📥

The biggest time-saver. First connect the official Gmail connector (Customize, then Connectors, log in with your own Google account). Then, on demand, ask:

📥 the inbox triage 
Using my connected Gmail, triage my new emails into urgent, needs reply, fyi, and junk,
apply those labels, and draft replies in my tone for the urgent and needs-reply ones.
Draft only. Do not send.
🔒 The safety part is built in: the Gmail connector can create drafts but physically cannot send (there is no send tool), so you always hit send yourself. It also only runs when you ask, there is no background bot watching your inbox. Sorting and tone-matching are best-effort, so skim before you send.
put it to work

Prompt 4: give it a to-do runner ✅

A little task board that starts clearing itself, with you watching.

✅ the task runner 
Build a small local to-do dashboard (a web UI on localhost) where I add, edit, and delete
tasks with priority levels, saved to tasks.json. Then work through the highest-priority
tasks I approve, log what you did to tasks.log, and mark them complete.
Honest note: keep it pointed at scoped, safe tasks and review the output. "Works through your list" means with your approval, not unattended on your whole backlog.
hands off

How to run these on a schedule (optional) ⏰

Two real ways, if you want prompts 2-4 to run without you kicking them off each time:

☁️ Routines Routines: cloud-scheduled, run with your laptop closed (Pro/Max/Team/Enterprise, preview).
💻 A cron job that calls Claude in headless mode (claude -p "...") on your own machine while it's on.
why it matters

The mindset shift 💡

Most people ask Claude one question at a time. That's like hiring a brilliant teammate and only letting them Google things for you. A memory so it knows you, a log so it keeps you honest, an inbox helper, and a task runner: same tool, completely different leverage.

Every link in one place 🔗

Want more like this? Follow @cindiezhu for daily AI you can actually use.