0:00 / 0:00

Paste This Into Claude, Never Hit a Token Limit Again

Name: Paste This Into Claude, Never Hit a Token Limit Again
Uploaded: 2026-08-29T15:45:43.000Z
Duration: 16 min 55 s
Channel: Jack Roberts
Description: A five-level system to avoid Claude token limits by cleaning up context, adjusting habits, improving communication, routing tasks to different models, and saving styles as skills.

Jack Roberts29 August 2026Watch on YouTube

Part of series

Ep. 14 · Obsidian opruimen met Claude AI

Stap-voor-stap uitleg hoe Claude AI Obsidian-notities automatisch herstructureert via het PARA-framework.

View the series

What you'll learn

An audit prompt reveals where tokens leak away, down to Jack's own example of 95.4% waste and three open sessions using 97.8% of tokens.
Using /clear between jobs, one model per session and batched questions stops Claude from rereading the entire history time and again.
The verbosity problem is solved with the C100 style, short sentences that limit follow-up questions and therefore token use.
Every screenshot costs 5,000 tokens, so codify designs you like into a skill or a spec of around 2 kilobytes.
Route work across models, with GraphiPy for codebases so Claude works from a map instead of the full files.

Frequently asked questions

How can an audit prompt lower my token usage in Claude?

The prompt lists everything loaded into a fresh session, such as system tools, MCP servers and Claude.md files, and flags files above 5K or totalling above 10K. In Jack's own case it showed 95.4% waste, of which 97.8% came from three open sessions.

Why is one model per session better for your token budget?

Switching models mid-conversation makes Claude reprocess the entire history. Keeping one model per session avoids that reload.

What is the verbosity problem and how does C100 solve it?

Some of Claude's answers are so long and complex that you keep asking clarifying questions, each costing tokens. C100 is a communication style in short sentences that limits the number of follow-up questions.

How many tokens does a screenshot cost and what is the alternative?

Every screenshot costs 5,000 tokens. Turning designs you like into a skill or a spec of about 2 kilobytes avoids repeatedly sending new images.

Topics

claude token limits prompt engineering ai tools productivity

Sources

What is known about this topic outside the broadcast, and where it says so.

Description from the channel

📈 GenSpark: https://bit.ly/49UHpPF 👉 Claude Prompt: https://bit.ly/46wn01a 🩵 Community: https://bit.ly/4kol0y5 *👋 Howdy* Hey, I'm Jack, I built and sold my last tech startup with a Gazillion customers, now I'm building AI startups and I share stuff that works. I quit my corporate job to do my start-up and have never looked back since. *💬 Overview* Claude token limits stop being a problem once you clean up what gets preloaded into every single session. This one walks through five levels that build on each other. Level one is a master prompt that audits your setup, and mine found 95.4% waste from open sessions and unused connectors. Level two covers the habits: /clear between jobs, one model per session, batching your questions, editing messages instead of correcting them, and text over PDFs. Level three is a D-100 communication skill so Claude's replies stay short and readable. Level four routes work to the right model, Haiku for grunt work, Sonnet for building, Opus and Fable for the big decisions, with Codex double checking Claude's work. Level five saves your design styles as skills, and Graphify gives Claude a map of any codebase. If you use Claude every day, this pays for itself fast. *Links* ☁️ Claude: https://claude.ai/ 🔥 Glaido: https://bit.ly/4eGoI3R (use code: WHSAAKXO for 1 FREE month) 🚀 GenSpark: https://www.genspark.ai/ 💬 ChatGPT: https://chatgpt.com/ 🗺️ Graphify 📁 GitHub: https://github.com/ 👾 Canva: https://canva.com ⚡ Hermes: https://nousresearch.com/ ⌚️ *Stamps:* 0:00 - Triple Your Claude Usage 0:33 - The Full System Overview 0:55 - Level 1 1:00 - Audit Your Token Waste 1:53 - Shocking Waste Revealed 2:52 - Cut Dead Connectors 3:38 - Sessions Draining Your Tokens 4:08 - Level 2 4:18 - Costly Habits To Stop 4:27 - Clear Between Jobs 5:10 - One Model Per Session 6:17 - Edit Don't Replace 7:07 - Text Beats PDFs 7:28 - Lock Your Effort Level 7:50 - Level 3 8:03 - Clearer Responses, Fewer Questions 8:44 - Claude Speaks Human 9:38 - Claude's Limits Exposed 10:08 - Any Model, One Platform 10:43 - AI Sheets In Seconds 11:06 - Instant Pitch Deck From Transcript 12:34 - Level 4 12:54 - Right Model, Right Job 13:27 - Use Codex To Check Work 14:17 - Stop Burning Tokens On Code 14:36 - Graphify Maps Your Codebase 15:34 - Level 5 15:52 - Screenshots Cost A Fortune 16:22 - Design Loop Outputs Winners 16:44 - What's Next If you keep hitting the Claude token limit or rate limit, this covers how to save Claude tokens, clean up your context window, use /clear, route work between Haiku, Sonnet, Opus and Fable, check code with Codex, and query codebases with Graphify. #Claude #ClaudeCode #TokenLimit #Anthropic #AITools #GenSpark