CLAUDE CODE AGENTS: FUNDAMENTALS - Chapter 10, Exercise 2
Adding a Search Breadth Signal to "Find the Caching Logic"
====================================================================================

QUESTION:
A brief simply says "find the caching logic," with no stated search
breadth. Explain what could go wrong in either direction (too shallow or
too wide a search), and rewrite the brief to include an appropriate breadth
signal.

SOLUTION / EXPLANATION:

WHAT GOES WRONG TOO SHALLOW:
If the research agent treats this as a quick lookup and stops at the first
plausible-looking match - say, a single file literally named "cache.ts" -
it might miss caching logic implemented elsewhere under a different name
or spread across multiple locations (an HTTP-layer cache, a database-query
cache, and a client-side cache might all exist separately, none of them
obviously named "cache"). The report would sound confident and complete
while actually missing most of the real picture.

WHAT GOES WRONG TOO WIDE:
If the research agent instead treats this as requiring an exhaustive,
thorough search across the entire codebase's every naming convention, and
the real answer was actually sitting in one obvious, easily found location,
the agent spends unnecessary effort exploring alternatives that were never
genuinely in doubt - a waste of time and search effort disproportionate to
how straightforward the actual question was.

Since the original brief gives no signal about which situation applies, the
agent has to guess, and could reasonably land on either extreme.

A well-formed rewrite, assuming (for this example) genuine uncertainty about
where caching might live:

"Find where caching is implemented across this codebase - I'm not sure if
it's centralized in one place or spread across multiple layers (HTTP
responses, database queries, computed values). A thorough search across
different naming conventions would be appropriate here, since I genuinely
don't know where to expect it."

This explicitly names the uncertainty driving the request and states the
appropriate breadth (thorough), giving the agent a clear calibration signal
instead of leaving it to guess.

--------------------------------------------------------------------------
WHY THIS WORKS AS AN ANSWER:
It explains concretely what could go wrong at each extreme (a false sense of
completeness from stopping early, versus wasted effort from over-searching
an easy question), and produces a rewrite that states genuine uncertainty
plus an explicit breadth signal, matching this chapter's own stated
guidance.