In Claude, open Settings, then Capabilities, and turn on code execution.
Go to Customize, then Skills, click +, then Create skill, then Upload a skill.
Upload the zip. Done.
Then just say one of these in any chat:
handoff moves you to a fresh chat without losing anything. Long chats are the biggest drain, because every message resends the whole chat.
limit check shows what is eating your limit in this chat and how to fix it.
lean mode keeps every answer short and cheap.
02
ChatGPT, or both
The paste-in version
ChatGPT does not take skill uploads, so paste this into Settings, then Personalization, then Custom instructions. It also works in Claude: Settings, then Instructions for Claude.
Paste once
Save my usage. Follow these on every reply:
1. Answer first. No intro, no repeating my question, no recap at the end.
2. Match the length to the question. A short question gets a short answer. Only go long when I ask for depth.
3. When you change something I gave you (text, code, a list), show only the changed parts, not the whole thing again. Give me the full version only if I say "full version".
4. If my request is unclear or missing something you need, ask me up to 2 short questions before you start. Do not guess and write a long answer I will have to redo.
5. Do not search the web, run code, or create files unless I ask, or the task cannot be done without it.
6. One option, not five. Give me your best pick and one line on why. I will ask if I want alternatives.
7. If this chat gets long (around 20 messages) or the topic changes, tell me in one line: "This chat is getting long. Say handoff and I will write you a brief for a fresh one."
When a ChatGPT chat gets long, paste this to move to a fresh one:
ChatGPT handoff
Write a handoff brief so I can continue this work in a brand new chat. Read our whole conversation. Do not ask me questions. Output ONLY the brief in one code block, under 250 words, in this exact shape:
# Handoff: <goal in one line>
## Where we are
- Done: <what is finished and agreed>
- In progress: <what we were in the middle of>
## Decisions (do not reopen these)
- <decision> because <why>
## Key details to keep
- <names, numbers, links, exact wording, preferences that would otherwise be lost>
## Latest version
<paste the current draft / plan / code only if it is short. If it is long, tell me which message to paste in after this brief.>
## Next step
<the single most immediate thing to do>
## To resume
Read this brief, confirm you understand in one line, then do the next step.
Rules: record decisions and why, not the story of how we got there. One next step, not a plan. Assume the reader knows nothing. Never include passwords or API keys.
Drop your email and the download unlocks right here. Plus every other resource in the library.
Keep going
More ways to get more out of the plan you pay for.
Shorter answers is one lever. These are the others: the Claude Code token savers, the model pairing that keeps your limits alive, and a cheaper worker model.