What in a pull request shows the author never read their own AI-written code?

The clearest signs are things a person who read the diff would have removed: comments that narrate the obvious, a helper written for one call site, error handling for cases that cannot happen, a test that asserts the mock it just set up, and files touched that have nothing to do with the ticket. Another giveaway is the description: it lists what the code does line by line but cannot say why this approach was chosen.

The quickest check in review is to ask one specific question about a decision in the diff. Someone who read it answers in a sentence; someone who did not goes back to the agent and pastes its answer. You do not need a policy for that, it shows up on its own.

To stop it earlier, move the first review before the pull request. Have the agent hand its work to a separate reviewer pass, require that tests and checks are actually run with the output shown, and keep pull requests small enough that the author can read the whole diff in a few minutes. Large agent-written diffs are where nobody reads anything.

Finally, make the author state in the description what they checked by hand. It is a small step, but it puts the responsibility back on a person rather than on the tool.

Listings mentioned

• Requesting code review · skill by obra
  Sends finished work to a separate reviewer subagent before it goes out, and fixes critical issues first.
• Verification before completion · skill by obra
  Makes the agent run the checks and show the output before it is allowed to say the work is done.
• Code Review Checklist Skill · skill by mohitagw15856
  Builds a review checklist for a pull request based on its language, type of change and risk level.

Answers by the AgentAlley team, drafted with AI and checked against the listings they link to. Not a real-person reply from the original thread.