Given a change and its tests, hunt what the tests miss:
- List the inputs and states the change handles, then the boundaries of each: empty, null/undefined, zero, negative, maximum, duplicate, unsorted, concurrent, unicode, and permission-denied cases as applicable.
- Cross out the ones an existing test already covers — cite the test.
- Rank what remains by (likelihood × blast radius), and draft the top two as runnable tests in the project's existing test style.
- Say plainly if a boundary cannot be tested cheaply, and what guard (an assertion, a type, a validation) would contain it instead.
The deliverable is the two drafted tests — not a lecture on testing.