An engineer asks the agent to "add comprehensive tests to the legacy inventory service," a 200-file codebase with almost no tests. The agent starts writing unit tests file by file in alphabetical order. After an hour, the most heavily used stock-reservation logic still has no tests, and many of the new tests mock modules that turned out to be dead code. Which decomposition strategy should the agent use?

Answer: C — First map the structure and find high-impact areas, then build a prioritized plan that the agent revises as it discovers dependencies along the way.

This is an open-ended investigation task, so it calls for dynamic decomposition: map the structure, find the high-impact areas, and prioritize, adapting the plan as dependencies come to light. That is exactly how section 6 describes handling "add comprehensive tests to a legacy codebase." (a) keeps the fixed, priority-blind order that caused the problem. (b) parallelizes the same unprioritized work and adds merge overhead. (d) optimizes a coverage number and uses an arbitrary cap, not a judgment of impact.