Ship Your First Change with Claude Code
Drive one real change end to end in Claude Code: pull a tiny journal app with a stubbed search feature and two failing tests, hand the agent the goal, and watch it gather, act, and verify until the suite is green. A local self-check proves the run happened for real, and you submit the code, the machine-generated evidence, and a short trace of the loop for instant, rubric-based feedback.
45 min
Est. time
3
Outcomes
5
Rubric criteria
55%
Pass score
What you'll learn
Skills you'll have real reps in after shipping this.
The scenario
You just installed Claude Code and you want to feel the loop for yourself instead of reading about it. The fastest way to internalize an agent is to hand it a small, self-contained goal that ends in a verifiable change, then watch it gather context, act, and check its own work.
This kit is that goal, prepared for you. tinylog is a tiny journal CLI where add and list work and search is a stub that raises NotImplementedError; two red tests describe exactly what search should do. You open Claude Code in the folder, give it the goal, and let it drive. You write none of the implementation yourself: your job is to steer, watch the loop close, and capture where each phase happened. A local self-check re-runs the suite and writes your proof-of-work evidence only when the feature genuinely works.
Your role
You are a developer running Claude Code for the first time. Your deliverable is the feature it built under your direction (a working search plus six green tests), machine-generated evidence of the passing run, and a short trace showing the gather, act, and verify phases in your own session, packaged as a single submission.
Start the task to unlock the full brief
You'll get the step-by-step requirements, setup commands, the 5-criterion grading rubric, tips, and the ability to submit your solution for instant AI grading.
Free to start · submit when you're ready
Run your first real Claude Code session, hands-on
This is the fastest way to feel how Claude Code actually works: pull a tiny journal CLI with a stubbed-out search feature and two failing tests, open Claude Code in the folder, and hand it the goal. You watch the agent gather context, edit the code, and run the test suite, then you trace the gather, act, and verify phases in your own transcript, so the agent loop stops being a diagram and becomes something you have driven. A local self-check keeps the whole thing honest: it re-runs the suite, confirms the spec tests are untouched, and writes your proof-of-work evidence only when the feature genuinely works. It is the first boss project in Preporato's hands-on Claude Code course.