Project P12 — Your Prompt Library

“The wise lay up knowledge, but the mouth of a fool brings ruin near.” — Proverbs 10:14 (ESV)

Chapter: 12 — Building Your Prompt Library Due: End of Week 12 Submit: A link to a shared folder or document (Google Drive, OneDrive, Dropbox, or a single PDF) containing your prompt-library document, your rubric scoring, screenshots of test runs, and a short README reflection. No GitHub, no code — this is a professional document portfolio. See Appendix A for where to keep your library and how to submit. Allowed tools: Any consumer AI assistant reachable from a browser or phone (ChatGPT, Claude, Gemini, or a free tier of any) for drafting and testing your templates; any document tool for the library itself (Google Docs, Word, Notion, OneNote, or a Google Sheet). Everything free-tier is fine. AI policy: AI is your drafting and critique partner, not your judge. Use it to help write, improve, and stress-test your templates — that is exactly the skill. But the organization, the naming, the quality verdict, and the decision of what belongs in the library are yours. Your README carries one honest AI usage note. You choose the tool; you own the verdict.

The Setup

You are Marisol Reyes, the operations coordinator at Second Harvest Riverside, a regional food bank with four paid staff and two hundred volunteers. You are, in practice, six people. Before lunch on a normal Tuesday you have written a thank-you to a major donor, drafted a volunteer-shift reminder, summarized a twelve-page grant report for your executive director, turned a chaotic staff-meeting recording into action items, replied to an upset volunteer, and written a two-sentence social post about tomorrow’s mobile pantry. Every one of those you wrote from scratch, in a hurry, slightly worse than the last time you wrote the same kind of thing.

Six weeks ago you started using an AI assistant, and it helped — genuinely. But you noticed something that bothers you: you keep re-writing the same prompts. The good prompt you wrote last month to summarize a grant report for your director? Gone. You wrote a clumsier one this month and got a clumsier summary. You are re-solving solved problems every single week, and the food bank’s donors, volunteers, and board are getting the slightly-worse version each time.

So you are going to do what a professional does: build a prompt library. Ten to fifteen reusable templates for the work you actually repeat, organized by the roles you actually wear, named so you can find them at 4:55 on a Friday, and tested so you know they produce good work every time — not just the time you got lucky. You are laying up knowledge, so that your worst, busiest day gets the benefit of your best, least-rushed thinking.

Setup (the starter)

This chapter ships two starter artifacts in code/. They are plain-text worksheets — nothing to install, nothing to run:

• code/prompt-library-starter.txt — a starter library with the field structure (Name, Role/Task, When to use, Model hint, Variables, Template, Example, Quality, Version) and three worked example templates (a customer/donor reply, a meeting-notes → action-items transformer, and an audience-targeted summarizer). Copy it into your own document and build on it.
• code/prompt-quality-rubric.txt — the eight-criterion quality rubric and the three-runs-then-different-inputs testing protocol. This is how you prove a template is library-ready.

Paste both into your own working document (or a folder) and make them yours. Do not submit them unchanged.

Learning Targets

You will demonstrate that you can:

• Turn your own repeated, one-off prompts into reusable templates with well-designed {{variables}} — the right things baked in, the right things left as blanks.
• Organize and name a library by role and task so any template is findable in seconds.
• Evaluate prompt quality deliberately — testing for consistency across runs and correctness across inputs — instead of trusting a prompt because it worked once.
• Version and iterate templates responsibly, recording why a change was made.
• Keep the spine rule intact under the temptation of a polished library: the template makes asking fast; you still own the verdict on every output.

Normal Tier

Goal: Build a real, organized, tested prompt library of 10–15 templates for your own actual (or your fictional Marisol-style) work.

Required features

1. 10–15 templates, each written in the full field structure from the starter (Name, Role/Task, When to use, Model hint, Variables, Template, Example, Quality, Version). They must be for real, repeated work — not toys.
2. At least three distinct roles represented (e.g. donor relations, volunteer coordination, board/exec reporting, social media, internal admin), organized as a two-level outline (role → task) with a name index at the top of the document.
3. Well-designed variables. Every template uses {{placeholders}} for what changes and bakes in what stays the same. No over-variabled “blank forms,” no welded-to-one-situation one-offs.
4. Tested templates. Score at least five of your templates on code/prompt-quality-rubric.txt — three same-input runs plus one different-input run each — and record each score out of 16, the weakest criterion, and what you’d fix. Include screenshots of at least two real test runs.
5. Versioning. At least two templates carry a real version history — a v2 (or higher) with a date and a one-line why note.
6. Safety. No real PII, PHI, passwords, or confidential figures baked into any template. Include a short safety note stating that private details live only in the {{blanks}} at use time, per Appendix C.
7. README reflection (400–700 words): what you built, which roles it covers, where the AI helped you draft/improve templates, where it steered you wrong, and your honest AI usage note.

Normal-tier rubric (out of 100)

Medium Tier (+up to 25% extra credit)

Make it a team asset.

1. Team-ready rewrite. Take your personal library and make it usable by someone who is not you: fill every When to use field for real, rewrite any private-shorthand names, and add a short “how to use this library” note at the top for a new teammate.
2. Explicit guardrails. Add a library-wide safety section stating what may never be pasted into a consumer AI tool (client/donor PII, financials, anything under a confidentiality agreement) — the shared-library rules from §12.7 and Appendix C.
3. Name an owner. Add a one-paragraph “library governance” note: who owns adding, approving, and pruning templates, and how a teammate proposes a new one. A shared library without an owner rots — show you know that.

Hard Tier (+up to 25% additional extra credit)

This is the judgment no AI can make for you, and it is graded as such.

Write what-to-template.pdf (or .docx) — a one-page memo addressed to your executive director — that makes and defends a single, discriminating recommendation: of Second Harvest’s (or your own) recurring communications, which should be templated and reused, and which must stay a fresh, thought-through, human-written response every single time — never on template autopilot.

You must:

• Name at least one task that belongs in the library (repeatable, low-stakes-per-instance, benefits from consistency) and defend why.
• Name at least one task that must never be templated (a condolence note to a bereaved volunteer, a sensitive donor apology, a crisis message, a personnel matter) and defend why a template would do harm there.
• Tie the line explicitly to the spine rule — you choose the tool; you own the verdict — and to Proverbs 10:14: the wise lay up knowledge, but a fluent, fast, templated word sent without weight can bring ruin near.

A memo that says “template everything” or “template nothing” fails. The grade is in the discrimination — showing you can tell the work that should be systematized from the work that must stay personal, and defend the boundary. An AI can fill your library; it cannot tell you which of your neighbors deserves an un-templated word. That judgment is the assignment.

Submission

Put everything in one shared folder (Google Drive / OneDrive / Dropbox) or a single PDF portfolio, and submit the link per Appendix A:

project-12/
  prompt-library.docx        (or .pdf / a shared Google Doc / a Google Sheet)
  rubric-scores.xlsx         (or .txt — your five+ template scores out of 16)
  test-run-screenshots/      (at least two real runs in an AI tool)
  README.txt                 (or .docx — your 400–700-word reflection)
  what-to-template.pdf        # Hard tier only

Everything is a document or a screenshot — .docx, .pdf, .xlsx, .txt, images. There is no code. Use placeholder names/figures anywhere a real one would be confidential; never paste real donor, client, or personnel data into your submission or into a consumer AI tool.

Hints (Read Before You Begin)

• Start from what you already repeat. Don’t invent templates for work you never do. Spend ten minutes listing the messages/tasks you write again and again — those are your library, sitting in plain sight.
• Rescue, don’t reinvent. Your best templates are hiding in your chat history and sent folder (Exercise Rep 1). Dig them up first; templatize second.
• Test on purpose, not by vibes. “It looked good” is not a rubric score. Actually run each tested template three times and read whether the shape held. The screenshots keep you honest.
• Write the Quality field before you think you need it. It is your verification checklist. A template without one is a template that invites you to send unread.
• Keep the private stuff in the blanks. If you catch yourself pasting a real donor’s name into a saved template, stop — that name is a {{variable}}, not part of the asset.
• Don’t over-tool it. One document beats three evenings comparing apps. Build the library first; move it later if you outgrow the doc.

What Mastery Looks Like (Beyond the Rubric)

A mastered submission is one where I can open your library and immediately find any template, read a Quality field that shows you know exactly what to verify in that output, and see rubric scores that prove you tested rather than hoped. Your version notes read like the memory of real lessons (“added this because it kept inventing owners”). And your Hard-tier memo tells me something an AI never could: that you know which of your food bank’s neighbors deserves a human word that no template should ever touch. The library is table stakes. The testing, the versioning, and the judgment about what not to template are the project.

Coach’s Note — The temptation this week is to build an impressive-looking library and never test it — to mistake polish for quality. Resist it. A pretty library of untested templates is exactly the trap of the chapter: it feels authoritative, so you stop reading the output. I would rather see eight templates you genuinely tested and can defend than fifteen you never ran. Depth over shine. Prove it works.

When You’re Done (a short checklist)

• 10–15 templates, every field filled, including a real Quality checklist.
• At least three roles, organized role→task, with a name index at the top.
• Variables designed on purpose — nothing over-variabled, nothing welded to one case.
• At least five templates scored on the rubric, with screenshots of real runs.
• At least two templates versioned with dates and why-notes.
• No confidential data baked into any template; safety note present.
• README written, honest, with an AI usage note.
• (Hard) what-to-template memo defends what to systematize and what to keep human.
• No real donor/client/personnel data anywhere in the submission.

A theological footnote. “The wise lay up knowledge, but the mouth of a fool brings ruin near” (Proverbs 10:14, ESV). This project is that proverb made into a Tuesday. Laying up knowledge is the humble, patient work of saving your best prompts where your worst day can reach them — refusing to let hard-won craft evaporate the moment you use it. But the verse has a second half, and your Hard-tier memo lives there. A library makes your words fast and fluent and constant, and that very fluency is what folly counterfeits — much said, quickly, confidently, without weight. The wisdom is not in how fast you can send; it is in what you choose to send, and in the neighbor you refuse to hand a templated word when they deserve a human one. Fill the storehouse. Then guard your mouth. Serve the donor, the volunteer, the grieving, and the board with work that is both prepared and weighed — because they are the neighbor this vocation was given to serve, and a fast wrong word still brings ruin near.

See you next week.