AI Guides › Claude Mastery
Building A Reusable Prompt Library Instead Of Reinventing Every Time
By Nigel Guy · 2 min read
Most people write a genuinely good prompt once — the wording that finally
got the tone right, the exact framing that produced a useful analysis — and
then never see it again. It's buried somewhere in a conversation from three
weeks ago, and next time the same task comes up, they start from a blank box
and reinvent it, usually slightly worse, because reconstructing something
from memory is harder than it looks.
The rule: a prompt that worked well once is worth saving deliberately, as
a reusable piece of infrastructure — not left to be rediscovered by accident
in old conversation history.
The mechanism: the Prompt Library
- Collect, don't compose from scratch. Whenever a prompt produces
something genuinely good — not just adequate — save the exact wording
somewhere outside the conversation: a note, a document, a simple file.
The instinct to move on to the next task is exactly what causes these to
get lost.
- Generalise it slightly as you save it. Replace the specific details
("Tuesday's board deck") with a placeholder ("[document name]") so the
prompt is reusable for the next instance of the same task, not just a
record of one past use.
- Name it by the job, not the date. "Weekly report first draft" is
findable later; "prompt from March 14" is not. The organising principle
is the recurring task, not the chronology of when you happened to write
it.
- Note what made it work, briefly. A one-line reason ("works because it
specifies the audience before the format") turns the entry into
something you can adapt deliberately, rather than a fixed incantation
you're afraid to touch.
- Revisit and prune periodically. A library that only grows becomes as
hard to search as no library at all — retire prompts for tasks you no
longer do, and update ones that have been superseded by a better
version.
What to skip
Skip building an elaborate categorisation system before you have more than
a handful of entries — a flat, well-named list is genuinely easier to use
early on than a folder structure designed for a library ten times its
current size. And skip saving every prompt indiscriminately; save the ones
that solved something, not every variant you tried along the way.
Guardrails
- A saved prompt encodes a good approach to a task, not a guarantee of a
good output every time — the same prompt can still need adjusting for a
new situation, and treating it as fully automatic invites exactly the
kind of unchecked output this whole library is meant to avoid.
- Don't save prompts containing sensitive specifics (real client names,
actual figures, personal data) into a shared or exported library — save
the generalised structure, not the instance that happened to contain
someone's private information.
- Prompts that reference specific product features or model behaviour can
go stale as the tool changes — a periodic prune, as above, is what keeps
the library trustworthy rather than a museum of old workarounds.
All 751 AI guides · JulieMango plans from £17/mo