About
Why this library exists.
This started as a prompt log, the working notebook a prompt engineer keeps to track and improve their craft. It grew into something I actually use: a working library of 427 prompts I reach for in real work.
The MASTER framework is what made the difference. The earlier version of my notebook was a pile of prompts that mostly worked. The current version is a pile of prompts that work the same way every time, fail in diagnosable ways, and can be upgraded with a single-variable change instead of a rewrite.
Principles
A prompt earns its place by being meaningfully different.
Not by being a rephrase of an adjacent prompt. The library has been audited three times for duplication; only templates with a distinct trigger condition and a distinct mechanism survived.
Full MASTER means all nine layers.
If a prompt is tagged Full MASTER, it has Mission, Agent, Situation, Task, Examples (where needed), Rules, Think, Eval, and Output. If any layer is missing, the tag is wrong, and the framework audit catches it.
The personal voice is the point.
The notes are written for future me, not for an audience. Terse, opinionated, occasionally wrong. The Notes field on each template is what makes the library teach me when I revisit it.
The bracketed placeholders are intentional.
In Part II, every template has [BRACKETED_VARIABLES] for the user to fill in. That is not laziness, it is design. A template that pre-fills the variables is a one-shot prompt, not a template.
Banned AI-tell phrases are banned for a reason.
'Delve,' 'in conclusion,' 'passionate about,' 'results-driven', these surface signals of pattern-completion without grounded reasoning. The RULES section of every prompt prohibits them. When the model defaults to them, it means the model isn't actually engaging with the situation.
How the library is organised
Three parts, each with a different purpose:
55 completed prompt engineering exercises, against real situations.
372 reusable templates across 18 categories.
A live feed of curated prompts from across the web.
Maintenance
On the first Friday of each quarter, I run the maintenance pass: re-test the most-used templates, run the constraint audit on the one that drifted most, retire any template I haven't used in 90 days. The library is durable; individual prompts are not.
Stack
Static Next.js site with Tailwind, deployed via GitHub Pages or Vercel. Search is client-side via Fuse.js, no analytics, no tracking, no third-party fonts beyond Google's Inter and JetBrains Mono. The whole site is under 500 KB.
Want to use the prompts?
Open the library, find what fits, fill the brackets, run it.