Capstone: build your personal ChatGPT toolkit
Bring everything together: a rules sheet, prompt library, test set, facts file, workflow and review log, documented in a README you can show.
In this lesson you will
- Choose one valuable recurring job
- Build a tested prompt library
- Document rules and a workflow
- Deliver a README as proof of skill
About 20 minutes
You are now the person who builds the system
Twenty-nine lessons ago you were typing questions into a box. Now you know how to prompt with precision, ask for structure, chain steps, supervise agents, automate with safety, read an API call, measure quality, work professionally and run projects. Dozens of people know a few of those things. Almost nobody has them organised into a personal system. This capstone lesson has one job: guide you to build that system, a working ChatGPT toolkit that is yours, documented and ready to use tomorrow. Treat it as a project, using lesson 29’s method.
What your finished toolkit contains
| Part | What it is | Where it lives |
|---|---|---|
| 1 Rules sheet | One page: approved tools, data traffic light, review checklist, disclosure habits | A document you can open in 5 seconds |
| 2 Prompt library | 10 to 20 tested prompts with names, purposes and fill-in fields | A spreadsheet or notes app |
| 3 Test set | 20 examples for your most important prompt, with expected answers | A spreadsheet |
| 4 Facts file | Confirmed information about you, your work or business that you paste into prompts | A document, kept up to date |
| 5 Workflow | One end-to-end chain for a task you do often, with human checkpoints | A written procedure |
| 6 Review log | Failures, fixes and time saved | A running list |
Step 1: choose the job
Pick one recurring task that matters: weekly reports, customer replies, content planning, lesson preparation, job applications, research summaries. You will build the toolkit around it. Write the goal in one sentence with a measurable result, for example “cut my weekly report from 90 to 30 minutes with no factual errors”.
Step 2: write the rules sheet
Interview me with up to 6 questions about my work and the data I handle. Then draft a one-page AI rules sheet containing: approved tools, what I never paste, my review checklist before sending anything, when I must disclose AI use, and who to ask when unsure. Do not invent policies; mark anything I need to confirm with my employer or school as CONFIRM.
Step 3: build the prompt library
A library entry has a name, a purpose, a template with fields and a note on what to check. Templates make good prompts repeatable. Here is a tiny library in code, with a function that fills the templates and warns about missing fields. It was run for real:
import string
LIBRARY = {
"weekly-report": {
"purpose": "Turn rough notes into a status report",
"template": "Write a status report from these notes. Sections: Done, In progress, Blocked, Next week. Under $words words. Use only facts in the notes.\n\nNotes:\n$notes",
"fields": ["words", "notes"],
},
"polite-decline": {
"purpose": "Decline a request kindly",
"template": "Write a polite, firm decline to this request: $request. Tone: $tone. Under 80 words. Offer one alternative: $alternative.",
"fields": ["request", "tone", "alternative"],
},
}
def render(name, **values):
entry = LIBRARY[name]
missing = [f for f in entry["fields"] if f not in values]
if missing:
return f"Cannot use '{name}': missing {', '.join(missing)}"
return string.Template(entry["template"]).substitute(values)
print(render("polite-decline", request="a free sample box", tone="warm",
alternative="a 10 percent discount code"))
print()
print(render("weekly-report", words="150"))
print()
for name, e in LIBRARY.items():
print(f"{name:<15} {e['purpose']}")
Write a polite, firm decline to this request: a free sample box. Tone: warm. Under 80 words. Offer one alternative: a 10 percent discount code.
Cannot use 'weekly-report': missing notes
weekly-report Turn rough notes into a status report
polite-decline Decline a request kindly
The second call is deliberately incomplete: the program refuses to build a prompt with missing information instead of sending a vague request. You can do exactly the same by hand: a good template has visible blanks that you must fill.
Step 4: test the key prompt
- Take your most important template.
- Collect 20 examples with expected outcomes (lesson 27).
- Run the prompt, score it and list the failures.
- Change one thing at a time, re-run and keep what improves the score.
- Record the final version, the score and the date in your library.
Step 5: design the workflow
Write your chain (lesson 23) as a numbered procedure: input, stages, outputs and where you check. Include a stop rule: when the output is unusual, a person decides. Example for customer replies: classify, extract facts, draft reply from the facts file, check against tone rules, you approve, send.
Step 6: close the loop
- Add each new failure to the test set.
- Review the library monthly: delete unused prompts, improve weak ones.
- Re-read the rules sheet when your tools or your job change.
- Record time saved, so you know the real payoff.
Your capstone deliverable
Put the six parts into one folder and write a half-page README explaining how you use it. Then do one real task with the toolkit and note three things: time taken, errors found and what you would improve. This is your proof of skill. It is also excellent material for a portfolio, a job interview or a freelance pitch.
MY CHATGPT TOOLKIT - README
Owner: [name] Updated: [date]
Main job: [one sentence with a measurable goal]
Files: 1 rules sheet | 2 prompt library | 3 test set | 4 facts file | 5 workflow | 6 review log
How I use it: [3 to 5 steps]
Latest test score of my key prompt: [x of 20]
Time per task: before [..] -> now [..]
Known limits: [what I never delegate]
Next improvement: [one item]
What to learn next
| Direction | Where to go |
|---|---|
| Better prompts for other models | The Prompt engineering and Claude basics courses |
| Automation and agents | The AI automation and agents course |
| Code behind the scenes | The Python course |
| Using AI for money-making skills | Freelancing, Blogging, Digital marketing and Copywriting |
| Turning data into answers | Data analysis and Excel |
Building a giant toolkit you never open. Start with one job and one prompt that you really use. Add parts only when you feel the need.
Put it in your calendar. Prompts, tools and your own needs all change. A short, regular tune-up keeps the toolkit alive.
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 aim of the capstone?
- To organise your skills into a personal toolkit
- To memorise model names
- To delete old chats
- To buy a plan
A system turns knowledge into habit.
What should a prompt template contain?
- Hidden random words
- Visible fields to fill in
- No instructions
- Only a name
Blanks force the needed details.
How do you improve your key prompt?
- Change ten things at once
- Skip testing
- Ask someone else to guess
- Change one thing and re-run the test set
One change at a time shows what helped.
What belongs in the rules sheet?
- Passwords
- Approved tools, data rules and a review checklist
- Client secrets
- Only jokes
It keeps use safe and consistent.
What is a sensible next step after this course?
- Stop practising
- Collect more tools
- Share your key
- Do one real task with the toolkit and note the results
Real use gives real evidence.
Summary
- Organise your skills into a toolkit of six parts.
- Templates make good prompts repeatable.
- Test, change one thing, re-test.
- Keep reviewing; use it on a real task.
Questions
How long does the capstone take?
Plan a few hours spread over a week: one for rules and library, one for testing, one for the workflow and README.
Can I share my toolkit?
Yes, without confidential data or private facts. Prompts and structure are great to share.
What if my job has strict rules?
Follow them first. Build the toolkit inside the approved tools and ask what you are allowed to do.
