Beginner project: build your starter prompt pack
Build a pack of five tested prompt templates for your own weekly tasks, with test inputs, a score sheet, notes and a README.
In this lesson you will
- Choose five real tasks
- Write templates with blanks and rules
- Test each on three inputs
- Document the pack in a README
About 20 minutes
From knowing to having
Hira has read nine lessons. Tomorrow she will open an AI chat and, with a deadline looming, start typing from scratch again. The goal of this project is to prevent that. By the end you will have a small starter prompt pack: five tested prompts for tasks you actually do, saved in one document, each with a name, a template, an example input and notes on how it performed. It is your first real prompt library, and the best portfolio start you can make at this level.
What you will build
| Part | What it contains |
|---|---|
| 5 templates | Five prompts with role, task, context and format and blanks to fill |
| 3 test inputs each | Real or realistic inputs, anonymised |
| A score sheet | A simple table to rate each result 1 to 5 |
| Notes | What you changed and what worked |
| A README | Half a page on how to use the pack |
Step 1: choose five tasks
Pick things you do every week. Examples: replying to messages, summarising an article or notes, explaining a hard topic, planning a day or a study schedule, writing a short post, checking a draft for mistakes, preparing for an interview. Aim for variety: one writing task, one summary, one explain-and-learn, one planning task, one checking task.
Step 2: write each template
Use the four parts. Replace changing details with blanks in square brackets or dollar names. Add the finish line and a rule for missing information. Here is the structure, shown as real code that fills the blanks and warns when a value is missing. The output is real:
import string
PACK = {
"email-reply": string.Template("Role: a polite $voice writer.\nTask: write a reply to the message below.\nGoal: $goal\nFormat: under $words words, end with a clear next step.\n\nMessage:\n$message"),
"summary": string.Template("Summarise the text below in $n bullet points, each under $limit words. Use only facts in the text; write UNKNOWN if missing.\n\nText:\n$text"),
"study-helper": string.Template("Role: a patient tutor. Explain $topic for $level in 5 sentences, then ask me one question to check understanding."),
}
def make(name, **kw):
t = PACK[name]
try:
return t.substitute(kw)
except KeyError as e:
return f"Missing value for {e.args[0]} in '{name}'"
print(make("study-helper", topic="photosynthesis", level="a 12-year-old"))
print()
print(make("summary", n=3, limit=15, text="[article here]"))
print()
print(make("email-reply", voice="friendly", goal="confirm the meeting", words=80))
Role: a patient tutor. Explain photosynthesis for a 12-year-old in 5 sentences, then ask me one question to check understanding.
Summarise the text below in 3 bullet points, each under 15 words. Use only facts in the text; write UNKNOWN if missing.
Text:
[article here]
Missing value for message in 'email-reply'
You do not need to use code for your pack. A document or note with [BRACKETS] works exactly the same. The code shows how the idea scales when a team or an app fills templates automatically.
Step 3: test each template three times
- Fill the template with a real input and run it.
- Read the answer. Rate it 1 to 5 for: correct, useful, right format, right tone.
- Note the single biggest problem.
- Fix it with one change and run again.
- Repeat with two more different inputs. Make sure the template works on easy and tricky cases.
| Template | Scores | Main problem | Fix |
|---|---|---|---|
| email-reply | 4, 3, 4 | Too long | Word limit |
| summary | 5, 4, 3 | Made-up number | UNKNOWN rule |
| study-helper | 4, 5, 4 | Too advanced | Level added |
The scores above are an example layout, not real results. Your own scores will differ, and that is the point of testing.
Step 4: write the README
MY STARTER PROMPT PACK
Owner: [your name] Updated: [date]
What it is: five tested prompts for my weekly tasks.
How I use it: copy a template, fill the [BLANKS], run, read, check facts, adjust.
Rules I follow: no private data; I check important facts; I disclose AI help where required.
Templates: 1 email-reply | 2 summary | 3 study-helper | 4 planner | 5 draft-checker
Best result so far: [which template, what score]
Next improvement: [one item]
Quality check for your project
- Every template has a role or audience, a task verb, context blanks and a format line.
- Every template says what to do if information is missing.
- Each has been tested on three different inputs.
- No real private data is saved in the pack.
- You can explain, in one sentence, why each template is written the way it is.
Going further
Share one template with a friend or classmate and ask them to use it without your help. Watch where they get stuck, because that shows where your instructions are unclear. In the Intermediate lessons you will turn this pack into a more powerful library with step-by-step prompts, structured outputs and test sheets.
Saving a prompt after one lucky result. A lucky answer can hide a weak template. Test on at least three different inputs before you call it finished.
Use names like ‘reply-calmly’ or ‘explain-simply’ rather than ‘prompt1’. In a month you will find the right one in seconds.
Check what you learned
Five quick questions. Pick an answer and you will see straight away whether it is right and why.
What is the project’s main output?
- A starter pack of five tested prompts
- A single long prompt
- A video
- A password list
A small tested library.
How many test inputs should each template get?
- None
- At least three
- One lucky one
- A hundred
Three different inputs reveal weak spots.
What should every template include?
- A secret code
- Private data
- A long story
- A rule for missing information
It prevents invented details.
Why name templates by outcome?
- It is required
- They are easy to find later
- It saves electricity
- It hides them
Clear names speed up reuse.
Why ask a friend to try a template?
- To make it longer
- To avoid testing
- To share passwords
- To find unclear instructions
Their confusion shows gaps.
Summary
- A starter pack turns knowledge into reusable assets.
- Templates need role or audience, task, context blanks and format.
- Test on three different inputs and fix one thing at a time.
- Document and keep private data out.
Questions
Do I need coding for this project?
No. A document or note works. The code shown is an optional illustration.
What if my templates give poor results?
Read the answers, name the main problem, make one change and test again. Poor results are the raw material for improvement.
Can I use this in my portfolio?
Yes. A tested prompt pack with scores and notes shows real skill. Remove private information first.
