The Token Cheats Cheat Sheet | DOPE CEOs AI Bootcamp
DOPE CEOs · AI Bootcamp
Back to Resources
Session 06 · Resource 01 · Reference anytime

The Token Cheats Cheat Sheet.

Claude re-reads your entire conversation every time it replies. A 20-message session burns ~105,000 tokens — and one measured session found 98.5% of that went to re-reading history, only 1.5% to new output. Once you see the leak, the moves are obvious. This is every move on one page.

Five of the six core moves are free and take zero new tools. The sixth is just looking at a clock.
The 6 Token-Saving Moves
#MoveWhy it works
1Restart, don't follow upFresh session every ~20 messages. Every follow-up re-reads the whole history again.
2Batch into one message3 prompts = 3 full reloads. One message, three tasks = one reload.
3Plan in Chat, build in CoworkNail it in Chat first. Revising inside Cowork re-reads the whole folder every time.
4Match the modelSonnet/Haiku for quick work. Frees 30–70% of budget. (Full table below.)
5Keep context files smallCowork reads them before every task. 22k → under 2k = same signal, 10× less noise.
6Spread the 5-hr windowRolling window. Split morning/afternoon/evening so earlier usage rolls off.
The 9 Mechanical Cheat Codes

The 6 moves are what to do. These are why it works — so you can invent your own.

Verified = confirmed in Anthropic's tooling/docs   Translated = a power-user move in operator language   = non-negotiable

1 · Screenshot tax Verified

In Chrome, screenshots are sized image blocks that accumulate and re-send every turn. 18 of them burned 17% of a Max plan in 5 turns. Move: tell it to READ the page, not photograph it.

2 · Point, don't search Translated

Making Claude FIND something burns tokens reading everything. Move: name the exact file, paste the exact section. "Look at the pricing below" beats "find the pricing."

3 · Don't break the cache Verified

Claude caches the stable front of your chat. Switching models OR adding a tool mid-session shatters it and spikes cost. Move: pick model + tools BEFORE you start.

4 · Direct the compact Translated

Don't just "summarize." Direct what survives: "Keep the decisions and open questions, drop the back-and-forth." 10× more signal at the same cost.

5 · Mess goes in a side window Translated

Exploratory work pollutes a thread — Claude starts referencing its own dead ends. Move: run the messy part in a throwaway chat, bring back only the clean answer.

6 · Flag parallel Verified

When tasks don't depend on each other, say so. Move: "Do these three — they're independent." It handles them together instead of in sequence.

7 · Three "start over" moves Verified

New chat = total reset. Clear = wipe context, keep workspace. Compact = summarize + continue. Match the reset to the situation.

8 · Put a meter on it Verified

Free Chrome extensions (Claude Token Counter et al) show live token count, context %, and usage bars on claude.ai. You can't manage a leak you can't see.

9 · Score against a bar Verified

"Score this 1–100 on X, and if under 90, tell me why and rewrite." The number forces honesty. "Is this good?" gets flattery. (Three flavors below.)

Cheat Code 9 — The Three Flavors of "Score It"
RAW QUALITY: Rank this 1-100 on how tight the copy is. If it's under 90, tell me exactly what's weak and give me a v2. AGAINST A TARGET: Score this against [the goal / a competitor's version / the result I want]. How far off am I, and what closes the gap? MULTI-DIMENSION: Rate this on these levels, 1-10 each: voice match, clarity, conversion potential, originality. Find the ONE weakest axis and fix only that.
The MarkItDown Move — Stop Feeding It Messy PDFs

A raw PDF carries rendering noise Claude has to chew through. Clean markdown keeps only what matters — headings, tables, lists. Same document, up to 70% fewer tokens (a 70k-token PDF can drop to ~21k) and better output, because the structure survives. The tool is MarkItDown by Microsoft. Free.

Everyone · one click

The free "PDF to Markdown" app in the Microsoft Store (built on MarkItDown). Drop a file, get clean .md back. No terminal, no setup.

Desktop users · automatic

MarkItDown ships an MCP server. Wire it in as a connector and every file you upload auto-converts to markdown before Claude reads it.

There's also a command-line version (markitdown file.pdf -o file.md) for the technically inclined — but the Store app is the one for most people.

Two More Operator Moves

Ask it to shorten itself

Paste a long output back and ask: "This is taking up too many tokens — how do I shorten it without losing the meaning?" It'll trim its own bloat. Works on outputs, context files, and prompts alike.

Trim your scheduled tasks

Anything that runs daily runs its token cost daily. Paste out your scheduled-task prompts and trim them the same way — a bloated prompt on a timer is a leak that bills you every morning.

When It Goes Wrong — Name It, Move
Failure modeLooks likeThe move
DriftStops following rules ~20 msgs inRestart. Carry context via notes.
Context overloadSlows, thrashes, "autocompact" churnThread's too full. Fresh chat. Trim the input.
Template collapseEvery output sounds the sameNew chat. Re-anchor with a fresh example.
Over-softeningHedges, won't say the hard thingTighten Custom Instructions Part 2.
HallucinationConfident, specific, wrongAsk for sources. Web Search on. Cross-check.
The meta-skillThe pro doesn't fix failure modes faster. They don't hit them — they switched chats before the chat fell apart. You're cooked? Switch. The session notes ARE your progress.
Bake The Rules In — The Token-Discipline Block

Claude Code users drop these rules into a CLAUDE.md file so every session follows them. Your equivalent: paste this into your Cowork Project instructions (or a Project's custom instructions on claude.ai). Then it runs these habits for you instead of you remembering them.

TOKEN & CONTEXT DISCIPLINE Treat my context window as a limited resource. Follow these without being reminded: READING FILES - Only read the specific file or section you need for the task. Don't read a whole folder "to be safe." - When you only need part of a file, read that part, not all of it. STAYING ON TRACK - If you're unsure what I want, ask me ONE question before building. Don't guess and produce a large wrong output I have to redo. - If I tell you you're off track, stop immediately and re-align. Don't finish the wrong approach first. KEEPING CONTEXT CLEAN - When this task is basically done and we're moving to a new one, tell me: "This looks like a new task - want to start fresh?" - When we've made a lot of progress and the thread is getting long, tell me it's a good time to summarize and continue in a new chat, and remind me I can name what to protect. OUTPUT - Be concise. Give me the answer and the why, not restated context. - Don't repeat large blocks of code or text back to me unless I ask.
The translationSame discipline a developer bakes into CLAUDE.md, living where you actually work: your Project instructions. Set it once, every session in that Project inherits it.