AI tools for reviewing a small code refactor

Sample input

Use a synthetic Python function that loops over a list of records, skips rows without an email, strips surrounding spaces, lowercases the email and appends it only if not already in the output list. The contract preserves first-seen order; non-string emails are skipped. No network or external packages are needed.

Steps

1. Write the original function locally and save its behavior before using any selected tool. Prepare rows: Alice@example.com, spaced alice@example.com, missing email, numeric email 7, bob@example.com, and BOB@example.com.
2. Record the tool, editor or chat surface, account tier and available allowance. For editor tools, confirm that this setup can run Python locally. If the required feature or quota is unavailable, stop instead of silently switching to paid access. Use only this fixture; repository-wide and cloud-agent work are separate evaluations.
3. Ask for a clearer implementation using a set for membership and a list for stable output order. State that missing and non-string values must be skipped, case is normalized, and unrelated code must not change.
4. Request only the function, a diff explanation and proposed tests. Save the assistant and account plan with its response.
5. Inspect for changed output ordering, extra dependencies and changed input mutation behavior. Execute both versions against the six-row fixture, an empty list, and a second input ordered as bob@example.com then Alice@example.com in a disposable local environment.
6. Keep a review table with input case, original output, proposed output, assertion and reviewer decision. The six-row fixture must produce alice@example.com then bob@example.com; empty input must produce an empty list. The bob-first input must return bob@example.com then alice@example.com, exposing any accidental alphabetical sort.

Passes if

The actual locally executed outputs match the contract, input records stay unchanged, the diff is confined to the function and the reviewer can explain the set/list distinction. Assistant-proposed test text is stored separately from actual execution output.

Fails if

The refactor sorts or reorders results, crashes on the numeric value, mutates input, adds dependencies, or claims tests passed without an execution record.