Stop rewriting the same prompt. A small, organised library pays back every week.
Most people write the same prompt four times before they notice. A library fixes that: a small set of reusable prompts, each with a clear job, a documented contract and a version history. The value is not the number of prompts; it is the number of times you stop solving the same problem.
Organise by job, not by model
Group prompts by the work they do: extract, summarise, rewrite, classify, draft, review. A model-specific folder goes stale the moment you switch tools. A job-based folder survives, because the job does not change when the model does. Within a group, name prompts by the decision they support — "support-ticket triage" rather than "prompt 14".
Write for the next person
A prompt in a library is read by someone who was not there when it was written: a teammate, a future you, or a buyer. That reader needs the variables, the output shape, one example of accepted output and the known limitations. If a prompt needs a five-minute explanation before use, it is not ready to be in the library; it is still a draft.
Keep the library small
A library of thirty prompts nobody uses is worse than five that are used daily. Add a prompt only when you have run it at least twice on real work, and retire one when it stops being used. The point is leverage, not collection. Review the list every quarter and delete or merge what has gone quiet — dead entries make the useful ones harder to find.
Example: a three-prompt library for a content team — "brief to outline", "outline to draft", "draft to house style". Each takes named variables, each returns a fixed shape, and together they cover most of the pipeline. Three prompts, one workflow, no duplication.
Share the constraints, not just the prompts
The most valuable artefact in a mature library is the house constraints file: banned words, required disclosures, tone rules, refusal rules. Prompts reference it instead of repeating it. When the policy changes, you update one file and every prompt inherits the change. This is how a library stays consistent as it grows, and it is the same reason a design system beats a folder of one-off styles.
Measure by time saved
Track roughly how long a task took before and after a prompt entered the library. That number is what justifies maintaining it and what tells you which prompts to improve. A prompt that saves twenty minutes a week is worth polishing; one that saves two minutes a month is noise, however elegant it looks.
Checklist
- Is the library organised by job, not by tool?
- Does every prompt carry its contract and one example?
- Did every entry earn its place with real use?
- Is there one shared constraints file instead of copy-paste?