A lot of viral Claude cheat codes are useless. These are the 4 I actually use every day to get better answers while burning way fewer tokens, so you stop running out halfway through.
If you use Claude Code, tokens are your fuel. Every message, every file Claude reads and every tool it runs uses some up. Run low and you hit your limit, or your bill climbs.
Here’s the one idea behind three of these hacks. Your context window is Claude’s short-term memory, and Claude Code resends the whole conversation with every request. When a chat feels slow or dumb, the context is usually full of junk, and you’re paying to haul that junk around on every message.
01
See what is filling it
Run: /context
It shows a colored grid of what’s filling your context right now. When a tool, memory file or skill is hogging space, it suggests a fix. Most people have never looked, so they don’t know a giant CLAUDE.md or a pile of unused connectors is riding along on every message.
Then fix what you see. Run /mcp and disable servers you aren’t using in this project. Trim your CLAUDE.md, since it loads at the start of every session. Anthropic suggests keeping it under 200 lines and moving workflow-specific instructions into skills, which only load when used.
02
See what is costing you
Run: /usage
On any paid plan it shows what’s counting against your limits. You get a split by skill, subagent, plugin and MCP server, plus recommendations like "Trim CLAUDE.md." When the footer says f to fix, press f and Claude applies the ones it can. (Anthropic’s cost guide)
01
Compact and say what to keep
Run: /compact
When the window is getting full (around halfway is a good habit), type /compact. It summarizes the conversation and keeps going from that summary, so you drop the back-and-forth but keep the decisions.
The move most people miss is telling it what to protect. Type /compact keep the final database schema and the brand color codes exactly as written and it focuses the summary on those details.
02
Clear for a new task
Run: /clear
Done with one thing and moving to something unrelated? Don’t keep typing in the same chat. /clear starts a fresh conversation with empty context. Compacting is itself a big request because it reads everything it summarizes, while clearing costs nothing. Rule of thumb: compact to keep going on the same task, clear to start a new one. (All commands)
01
Brains where it counts
Not every task needs the most powerful model. Put a top model like Opus 5.5 on the work that needs judgment: planning the approach, big decisions and reviewing code. Anthropic says Sonnet 5.5 handles most everyday coding well for less. Switch anytime with /model.
02
Haiku for the grunt work
Subagents, the helpers Claude sends off to read files, search and run tests, don’t need genius. They need speed. Add one line to a subagent’s definition file: model: haiku. Without it, a custom subagent runs on your main model, so if you’re on Opus, every helper is too (subagent docs).
Be Honest About Fable
Fable 5.1 is the premium option, not the money saver. Its API price is $10 in and $50 out per million tokens, which is 2.5 times Opus 5.5. Haiku 5.5 is $0.10 and $0.50 for prompts up to 100,000 tokens, about 40 times cheaper than Opus (pricing). Use Fable when you want maximum brainpower on the main task. The savings come from matching cheap tasks to the cheap model.
03
Lower the effort on easy stuff
Thinking tokens bill as output, and the 5.5 models always think. For simple tasks, type /effort low or /effort medium so Claude spends less reasoning on them.
01
Don’t sit and watch it burn
This is the most expensive mistake I see. Claude heads down the wrong path and people let it finish so they can correct it after. That’s double the cost: you pay for the wrong version, then pay again for the fix.
The second it goes sideways, hit Esc and redirect: "Stop, that’s not what I meant. I want X, not Y." If it already changed files, double-tap Esc or type /rewind to roll the code and chat back to a checkpoint. For anything big, press Shift+Tab into plan mode first so you approve the approach before it builds.
Bake these habits into Claude so you don’t have to remember them. Paste this into the CLAUDE.md file in your project’s main folder (or create one). New to .md files? They’re plain text notes Claude reads. Here’s my crash course. Claude Code also follows the Compact instructions section whenever it summarizes your chat.
Adjust any line to fit how you work. Then tokens stop being a mystery. You can see where they go, reset on purpose, put cheap work on the cheap model and stop paying for wrong turns.