CLAUDE CODE AGENTS: FUNDAMENTALS - Chapter 8, Exercise 2
Refactoring a Function With No Existing Tests
====================================================================================

QUESTION:
A refactoring agent is asked to simplify a function that has no existing
tests at all. Explain what it should do before making any structural
change, using this chapter's own guidance.

SOLUTION / EXPLANATION:

This chapter's own system-prompt guidance is explicit that a refactoring
agent should flag missing or inadequate test coverage rather than proceeding
as though the absence of failing tests proves nothing broke. With genuinely
zero existing tests for this function, there is no "before" baseline to
compare against at all - the entire before/after verification method this
chapter relies on has nothing to work with.

Per this chapter's own tip box, the right move is to check (and if
necessary establish) test coverage BEFORE refactoring, not after. Concretely,
that means the refactoring agent should flag the missing coverage rather
than proceeding immediately, and either the coverage gap gets addressed
first - potentially delegating that work to a testing agent (Chapter 7) to
write tests capturing the function's current behavior - or the refactor
proceeds with an explicit acknowledgment that there is no automated way to
confirm behavior was actually preserved, and extra manual review is needed
instead.

What the agent should NOT do is simplify the function and simply report
that no tests failed, since with zero tests to begin with, "no tests
failed" is a meaningless, guaranteed-true statement that provides no actual
evidence behavior was preserved - it isn't a genuine verification result at
all.

--------------------------------------------------------------------------
WHY THIS WORKS AS AN ANSWER:
It identifies specifically why zero existing tests removes the entire
before/after verification method this chapter depends on, and names the
concrete correct response (flag the gap, establish coverage first or
proceed with explicit manual-review caveats) rather than treating "no
tests failed" as meaningful verification when there were no tests to fail
in the first place.