Summary
The first serious habit of a working AI operator: a versioned, tested, documented library of prompts. Not a Notion doc of tricks — a repository, with tests, changelog, and a decision rule for retiring a prompt.
Objectives
- 01Assemble ten production prompts for tasks you actually do.
- 02Version each prompt with a changelog and a test suite.
- 03Publish the library so it can be reused by a teammate.
- 04Defend the design choices in an oral review.
Key Ideas
- A prompt without a test is folklore.
- Version prompts the way you version code — with a changelog.
Lab
Build a versioned prompt library of ten prompts for tasks in a domain you know well (writing, coding, analysis, research, or operations). Each prompt must include: intent, input schema, example input/output, evaluation criteria, and known failure modes. Store in a Git repository. Add a test file that runs each prompt against three example inputs and asserts the output shape and one substantive property. Write a README explaining the library's scope, versioning discipline, and how to add a new prompt.
Deliverables
- — Public Git repository with ten versioned prompts.
- — Automated test file that runs and reports pass/fail per prompt.
- — README with scope, contribution guide, and changelog.
- — One-page design memo defending prompt taxonomy and eval choices.
Rubric
- — Coverage — the ten prompts span meaningfully different tasks.
- — Rigor — every prompt has a test that would catch a regression.
- — Documentation — a stranger can add a prompt without asking.
- — Judgment — the design memo shows real thinking about tradeoffs.