104: Checking the Agent's Work

Overview

The first three chapters teach you to brief an agent and to run a session with it. This one teaches you to check what comes back. An agent's report on its own work comes from the party with the least reason to doubt it, and the failures in this chapter share one shape: the report looks exactly like a correct one. A test suite passes without testing the change. A search returns zero because it opened no files. A status reads green because a pipe swallowed the real exit code. A destructive step goes ahead because the warning on screen was read as permission.