Build & submit taskBetabeginner

Build a Prompt Pack That Survives Three Rounds of Iteration

Take three vague business requests, decompose each into a structured prompt, then iterate three rounds and capture the verbatim Claude output at every round. The deliverable is a reusable prompt pack with a change log that names the specific edit responsible for each quality gain, scored against criteria you set before you started. No coding required.

2 hrs

Est. time

4

Outcomes

8

Rubric criteria

65%

Pass score

What you'll learn

Skills you'll have real reps in after shipping this.

Decomposition is the whole technique
Naming the audience, source, question, format, and constraints turns a guess into a specification. Most of the quality gain arrives in this step, before any iteration.
Change one thing per round
If a round edits five things and the output improves, you have learned nothing you can reuse. Single-variable iteration is what turns a lucky prompt into a transferable one.
Criteria set in advance
Deciding what good looks like before you see the output stops you from grading the result against whatever it happened to produce.
Task type changes the strategy
A drafting request needs voice and format constraints, an analysis request needs the source and the decision it feeds, and a brainstorming request needs volume and a spread instruction. One prompt shape does not serve all four.

See how it works

Why specificity moves the output

Few-shot shifts the boundary
systemClassify the message as billing, account, or technical. Reply with only the label.
(no examples, the model only has the instruction)
I can't sign in even after resetting my password.technical ✗
The export button does nothing when I click it.technical
Cancel my subscription and refund the last charge.account ✗
Examples teach the convention. With no examples the model guesses "technical" for a login problem. A couple of labeled user→assistant turns pin the boundary (password → account), and the held-out cases snap to the right label.

A request that names the audience, the source, and the output shape narrows what Claude has to guess at. Most of the quality gain in this task comes from the decomposition step rather than from later rounds of tinkering.

The scenario

Your team asks Claude for "a summary of the customer feedback" and gets something generic every time, so everyone quietly rewrites the output by hand and concludes the tool is overrated. The prompt is the problem. A request that does not say who the reader is, what decision the output feeds, which source material counts, or what format the result has to land in leaves Claude guessing, and it guesses toward the safe middle.

Task decomposition is the fix the exam keeps returning to. Instead of one broad ask, you break a request into its parts (audience, source, the specific question, the output shape, the constraints) and state each one. This task makes you do that three times over on real business requests, then prove the decomposition worked by iterating and recording what changed. You are building a prompt pack, which is a document your colleagues can reuse, so the value has to be legible to somebody who was not in the room.

Your role

You are the person on the team who actually gets useful work out of Claude. Your deliverable is a reusable prompt pack for three recurring business requests, with verbatim before-and-after evidence and a change log that says which edit earned which improvement.

Start the task to unlock the full brief

You'll get the step-by-step requirements, setup commands, the 8-criterion grading rubric, tips, and the ability to submit your solution for instant AI grading.

Free to start · submit when you're ready

What you'll build in this Claude prompting task

This is a build-and-submit task rather than a guided lab, and it requires no coding at all. You take three vague business requests of the kind your team actually sends, decompose each one into a stated audience, source, question, format, and set of constraints, then iterate three rounds and paste the verbatim Claude output so the improvement is visible rather than claimed.

The discipline being trained is single-variable iteration. Changing one thing per round is what separates a prompt you can hand to a colleague from one that happened to work once. You set your quality criteria before running anything, score the first and final output against those same criteria, and close with a reuse note so the pack survives contact with somebody who was not in the room.

Grading is rubric-based and explainable. Your document is scored against weighted criteria covering decomposition, verbatim evidence, attributable edits, and the scorecard, with per-criterion feedback quoted from your submission. The pass threshold is 65 percent and you can resubmit. Prompting and task execution is a scored domain on the Claude Certified Associate Foundations exam.

Frequently asked questions

Do I need to write code or use the API?

No. Everything happens in claude.ai, including the free tier. The deliverable is a Markdown or PDF document, and the Claude Certified Associate Foundations certification is designed for professionals with no software development or API experience.

Why does the task insist on changing only one thing per iteration?

If a round edits the audience, the format, and the constraints all at once and the output improves, you cannot say which edit did it, so nothing transfers to the next prompt you write. Single-variable iteration is what turns a lucky result into a technique you can reuse.

What if my final output is no better than my first?

Report it honestly and diagnose why. A null result with a credible diagnosis scores well on this rubric. A fabricated improvement is easy to spot in the pasted transcripts and scores badly.

What counts as a complete submission?

One Markdown, text, or PDF file covering three business requests of different task types, each with a quoted vague opener, an explicit decomposition, three numbered iterations with verbatim Claude output for rounds 1 and 3, a change log naming one edit per round, and a scorecard applying criteria you declared up front.