Finish one useful loop.
Your first session does not need a large feature. Choose a small question about a real project and establish a result you can inspect. The aim is to learn how your assistant reads context, explains decisions, and reports evidence.
- 01
Read the relevant code
Ask where a single behavior is implemented and which test is closest to it. Require file paths and symbols. Keep the first request read-only, then compare the answer with the repository. If the explanation is wrong, correct the context before authorizing a change.
- 02
Agree on one change
Pick a bounded improvement, such as a missing empty-state message. State the existing behavior and the desired result. Identify the screen or module involved, then let the assistant inspect the local patterns. Avoid attaching a neighboring refactor to the first task.
- 03
Check and close
Run the relevant check and inspect the changed behavior. Review the diff for unrelated edits. Finish with a record of what changed, what was verified, and what remains open. A completed small loop is a reliable foundation for the next request.
A useful finish
You can explain the current behavior, identify the proposed change, and name the evidence that will establish success.
A first request
Read only. Locate the code for [small behavior] and the nearest relevant test. Explain the current behavior with file references. Suggest one bounded improvement and a way to verify it. Do not modify files yet.