Why These Three
Small moves, real leverage
Most people use Claude at maybe 30% of what it can do, and it's not because they're missing some secret feature. It's the small stuff. The habits nobody tells you because they sound too simple to matter.
These are three I'd hand anyone on day one. The first fixes the number one reason your outputs quietly get worse over a long chat. The second saves you tokens without you doing anything different. The third makes everything you write with AI stop sounding like AI. None of them take more than a couple minutes to set up, and you'll use all three every single day.
The name trick
a full-context warning light for free
Add one line to your Claude global instructions: tell it to call you by your name every time it replies.
Here's why that's more than a cute touch. Every chat has a context window, basically Claude's short-term memory for that conversation. The longer the chat runs, the fuller it gets, and once it's too full the quality of the responses starts sliding without any warning. No error, no popup. It just quietly gets dumber.
When you've told Claude to use your name every reply, you get a warning light. The moment it stops calling you by name is the moment the context is getting too full. That's your signal to start a fresh chat before the outputs go sideways. It's the cheapest reliability check you'll ever set up, and it runs on autopilot forever.
Paste this into your global instructions (in the Claude app: Settings, then Profile, then the personal preferences box. In Claude Code: your CLAUDE.md or global config). Swap in your actual name.
Add to your global instructions
Call me [YOUR NAME] every single time you reply. The moment you stop using my name, I'll know the context window is getting too full and I need to start a fresh chat. Treat it as a reliability signal, not a formality. A few standing rules while you're at it: - Be direct. No preamble, no "great question," no summarizing what I just said back to me. - If I'm about to do something dumb, tell me before you do it. - When you're not sure, say you're not sure. Don't guess and dress it up as fact.
Switch models by the job
stop paying premium prices for cheap work
Most people pick one model and run everything through it. That's like keeping a Formula 1 car idling in traffic. The move is to switch models based on what you're actually doing.
The rule I use: the smartest, most powerful model for thinking. A lighter model for doing.
Use a top-tier model like Fable for the hard thinking, planning a project, architecting a build, writing a skill file or an agent, breaking a big messy goal into steps. That's where raw reasoning earns its keep and you want the best brain in the room.
Then switch to Opus or Sonnet for execution, the actual typing-it-out work once the plan exists. Running every follow-up on the most expensive model burns through your tokens (and your usage limits) way faster than you need to. Match the model to the weight of the task and the same amount of work stretches a lot further.
How to actually switch
In Claude Code: type /model and pick from the list. You can swap mid-session, so plan on the powerful model, run /model, and drop to a lighter one to build.
In the Claude app: the model picker sits right under the message box. Tap it and choose before you send.
The cheat: plan on the smart one, build on the cheap one.
Ask the powerful model to write you a clear, step-by-step plan first. Then switch models and tell the lighter one to execute the plan exactly. You keep the quality of the thinking and pay execution prices for the execution. That one habit alone stretches your usage further than any other trick on this list.
Install stop-slop
by Hardik Pandya · hardikpandya/stop-slop
You can always tell when something was written by AI. The em dash every other sentence. "Here's the thing." "It's not just X, it's Y." The weirdly even rhythm where every sentence is the same length. Once you notice the tells you can't unsee them, and neither can your audience.
stop-slop is a free Claude skill that strips all of it out. It rewrites the slop into something that reads like a person wrote it, and it scores your draft so you actually know when it crossed over from robotic to human. If you publish anything you wrote with AI (posts, emails, newsletters, captions) run it through this first. It's one of the fastest ways to upgrade the quality of everything you put out.
It installs in one line through npx skills, the package manager for Claude skills. Paste the command in your terminal, or hand it to Claude Code and let it run for you. The -g installs it globally so it's there in every project.
Install it
npx skills add hardikpandya/stop-slop -g
Try this
Run stop-slop on this draft. Strip every AI tell, then give me the human score before and after: [paste your text]
Full credit to Hardik Pandya, who built and open-sourced stop-slop. The skill lives at github.com/hardikpandya/stop-slop. Free and open, go star it.
Quick Reference
All three, in order
1. The name trick
Tell Claude to use your name every reply. When it stops, your context is full. Start a fresh chat.
2. Switch models by the job
Powerful model to plan (Fable), lighter model to execute (Opus or Sonnet). Plan on the smart one, build on the cheap one.
3. Install stop-slop
npx skills add hardikpandya/stop-slop -g. Run every AI draft through it before you publish.
Work with Me
Want me on your problem for an hour?
Book a 1:1 call and we build the fix live — the workflow you want automated, the tool you can’t crack, whatever’s stuck. Direct, hands-on, no pitch waiting at the end.