What I check in the last 15 minutes before merging an agent-generated PR

The number one way that teams new to Agentic Engineering workflows accumulate long-term technical debt and scope drift is by merging agent PRs that look fine in the diff but break on edge cases: empty input, scope drift buried in a "refactor," and tests that passed without testing anything.

I wrote up the 15-minute pass I use before merge, including a worked example of what it catches on a synthetic agent PR. I just want to share in case others find it useful.