Your own prompt library in 30 min a week

A good prompt sitting in a chat history is gone the next time you need it. A library is small, named, and written so you can use it without the old chat. Thirty minutes a week is enough if you rescue one prompt and test it once. Collecting more than that builds a graveyard.

One card, six fields

Notes, a spreadsheet or a folder: every card has the same fields. If one is missing, the prompt is not usable the next time you open it.

• Name: the job, not the tool. “Offer from bullet points”, not “ChatGPT prompt 14”.
• When: one sentence on when you use it, and when you don’t.
• What you insert: the placeholders, for example bullets, tone, length.
• Prompt: the text, with placeholders in square brackets.
• Example: a shortened output that already worked. Ten lines at most.
• Breaks when: the one failure this prompt often has. “Invents prices if none are in the bullets.”

The weekly rhythm

• Minute 0 to 10: find the one chat from this week where you used the answer almost unchanged. Ignore the rest.
• Minute 10 to 20: fill the six fields. Rename anything you would not understand in a month.
• Minute 20 to 30: run the prompt on a new, real task. If it fails, add to “Breaks when” or change the prompt. Keep one version, not five.

After eight weeks you have eight cards you have actually used. That beats a pile of forty prompts from a thread you never rewrote in your own words.

Name:
Use when:
Do not use when:
Insert: [bullets] [tone] [length]
Prompt:
Example (shortened):
Breaks when:

File by job, not by model

Name the folders after the work: writing, offers, meetings, images. A prompt that runs in Claude and in ChatGPT lives there once, not twice under the tool name. If a prompt only holds in one model, write that under “Use when”. Choosing the model itself is covered in ChatGPT, Claude or Gemini — which for what?.

How building blocks become a system is in Prompt Systems. A large collection filed by task is The Large Prompt Library. Versions and tests, once the cards multiply, are in Prompt Studio. The five gaps where cards usually fail are in 5 prompt mistakes every beginner makes.