You commented PROMPT, so here's everything from the reel: the skill, the three leaks with the fix for each, and my master prompt. Nothing held back for a later issue.
Why you hit the limit
It's not how much you use Claude. It's how many times you pay for the same message. You type a rough prompt, Claude does exactly what you typed, then you spend three more messages explaining what you meant. Four messages for one job.
Fix the first message and you fix the other three.
Part 1: Prompt Master (free)
Prompt Master is a free, open-source Claude skill (MIT licence, about 15,000 stars on GitHub). You give it one messy sentence and it hands back the prompt you should have written: what to build, what Claude can touch, and when to stop and ask you.
It works for prompts aimed at other tools too: ChatGPT, Cursor, Claude Code, Midjourney and more.
Install it in about 30 seconds:
Go to github.com/nidhinjs/prompt-master, click the green Code button, then Download ZIP.
Open claude.ai. In the sidebar go to Customize, then Skills.
Click Upload a skill and pick the ZIP. Done.
Use it: start a message with /prompt-master and say what you want, badly. For example:
/prompt-master landing page for my coffee brand, modern vibe, sign up formRead the prompt it gives back, fix anything it got wrong, then run it. One message instead of four.
Part 2: The leak cheat sheet
Prompt Master fixes one leak. Anthropic's own help page on usage limits lists more things that eat your limit, including conversation length, file attachments and model choice. I call it the leak tax. Here are the three most people pay every day.
Leak 1: Long chats
Claude rereads the conversation as it goes, so the longer the thread, the more each new message costs.
Fix: start a new chat when the topic changes. If you need the old context, ask Claude for a five-line summary first and paste it into the new chat.
Leak 2: Files you keep re-uploading
Uploading the same brand guide, price list or codebase into every chat costs you every time.
Fix: put files you reuse in a Project. Anthropic says project content is cached and counts less against your limit when it's reused.
Leak 3: The biggest model for tiny jobs
Model choice is on Anthropic's list too. Running the top model to fix a typo or reformat a list burns limit you'll want later.
Fix: drop down a model for the small stuff and save the big one for real thinking.
Bonus: put related questions in one message instead of five. Anthropic's own guide says planning your conversation cuts the back-and-forth.
Part 3: My master prompt
This is the brief I start any real task with. Paste it, fill in the brackets, and Claude asks you the questions before it burns your messages guessing.
You're helping me with: [THE TASK, one sentence].
Goal: [what done looks like, e.g. a landing page I can publish today].
Audience: [who it's for].
What you can touch: [files, tools or sections you may change].
What you must not touch: [anything off limits].
Format: [how you want the answer, e.g. one code file, a table, 5 bullet points].
Before you start:
1. Ask me up to 3 questions if anything above is unclear. Ask them all in one message.
2. If nothing is unclear, say "No questions" and start.
While you work:
- Do exactly what's asked. Don't add extras I didn't ask for.
- If you hit a decision I didn't cover, pick the simplest option and tell me which one you picked.
- Stop and ask before anything you can't undo.
When you finish, give me the result first, then one line on anything I should check.It does the same job as Prompt Master, just by hand: it makes Claude ask everything up front, in one message, instead of guessing and making you correct it three times.
From Mike
That's the lot. Install the skill, fix the three leaks, keep the master prompt somewhere you can paste it from. Your limit will last a lot longer.
If you want AI set up properly inside your business, not just in your chat window, OptiMAX does it done-for-you. Email [email protected] and tell me what you're trying to automate.
Mike