Project 01 — Your First 10 Conversations

Enforces: the four prompt levers, iteration, file uploads, model choice (from claude-101) Surface: claude.ai (no code) · Time: ~45 min · Difficulty: 🟢 starter

Why this project

Reading about prompting doesn't transfer; doing it does. This project forces you to run ten real tasks and consciously apply the four levers (Role, Task, Context, Format), so the habit forms.

What you'll build

A short "prompt log" — a single document (use a claude.ai Artifact or any notes app) capturing ten conversations, plus one prompt you measurably improved.

Steps

Set up — Create a Project in claude.ai called "Learning Claude" and add custom instructions: "Explain your reasoning briefly and ask me a clarifying question when my request is vague."
Run ten tasks — Do at least one from each category below. For each, write the prompt using all four levers where they apply:
- Summarize: paste an article and get a 5-bullet summary for a specific audience.
- Draft: a short email or message with a clear goal and tone.
- Explain: a concept you don't understand, "to a 10-year-old."
- Plan: a step-by-step plan for a real task this week.
- Analyze a file: upload a PDF/spreadsheet and ask three grounded questions.
- Brainstorm: 10 ideas, then ask Claude to pick the best 3 and say why.
- Rewrite: paste your own writing and ask for a tighter version.
- Compare: ask Claude to weigh two options in a table.
- Roleplay: have Claude interview you to draft something (e.g., a bio).
- Critique: ask Claude to find weaknesses in a plan you wrote.
The improvement rep — Take your worst result. Diagnose which lever was missing, rewrite the prompt, and re-run. Save the before/after.
Model check — Re-run one simple task on a faster model (Haiku) and one hard task on Opus. Note the difference in your log.

Acceptance criteria — you're done when

You completed ten conversations across at least six categories.
At least one task used an uploaded file with grounded questions.
You have one documented before → after prompt improvement with the missing lever named.
Your prompt log notes one case where a different model tier was clearly better.
You journaled the single biggest thing that improved your results in learning-journal-template.

Stretch goals

Add a knowledge file to your Project and show an answer that depends on it.
Build a reusable checklist as an Artifact and iterate it twice.
Write a "house style" instruction and verify two chats obey it.

Self-assessment rubric

Level	Signal
🟢 Got it	You reach for the four levers automatically and iterate instead of restarting.
🟡 Almost	You write good prompts but still occasionally re-prompt from scratch.
🔴 Revisit	Answers are still hit-or-miss; re-watch claude-101 Module 2.