GUIDE-002 · CHECKED

AI code review before merge needs more than a green summary.

A useful review starts with two artifacts: the code change and the record of what the agent actually ran. Compare them, classify any failure or missing check, then make the merge decision with a person.

Audience
Reviewers receiving AI-generated pull requests
Checked
2026-11-19
Boundary
Not a correctness certificate

§ 01

Build the reviewer handoff

Start with the diff, the agent's summary and the agent run record. The diff tells you what changed. The run record tells you what the agent was asked to do, which tools ran, what outputs were recorded and which checks were skipped.

Keep the artifacts together in the review context. A reviewer should not have to infer whether "tests pass" means a real command exited successfully or only repeats the agent's claim.

§ 02

Classify failures before fixing them

  • Runner failure: the environment, tool call or command could not complete.
  • Candidate failure: a check ran and reported a problem in the generated change.
  • Verification gap: the relevant check never ran, so the claim remains unverified.

This classification prevents a failed runner from being mistaken for a bad patch, and prevents an unrun check from being reported as a pass. See the runner health guide for the environment side.

§ 03

Keep the merge decision human

Use the run record to narrow the questions, then read the changed code and the relevant test coverage yourself. A recorded command is evidence that the command ran; it is not a correctness certificate.

For a second layer, compare the change with the patch verification guide and keep the boundary explicit: review evidence supports a decision, but does not make the decision for you.

§ 04

FAQ

Q.01What should a reviewer inspect before merging AI code?

Inspect the diff, the agent's run record, the recorded check outputs and the checks that were skipped. Then decide with a person.

Q.02Is a green agent summary enough?

No. Compare the summary with recorded commands and outputs. A summary without a corresponding check stays a claim.

Q.03Does AdaptOrch replace code review?

No. AdaptOrch records the agent run so the reviewer can inspect it; the reviewer still evaluates the code and decides whether to merge.

NEXT

Try it on one agent change.

Connect your coding agent, run one change through it, and read the record next to your usual review. Starter is free.

START FREE ON STARTER

Starter is free with 50 admitted runs a month. Pro is $39 a month for 500 admitted runs and 5 workspace users. Team is $149 a month for 2,500 admitted runs and 15 workspace users. Compare plans · Start free on Starter

Connect from Claude Code, Codex CLI or Cursor · Read the AI PR review guide