Produce Review Packets

produce-review-packets

Summary

Reviewers lose time when a handoff contains only a diff and no acceptance evidence. Package the goal, validation, risks, and proof artifact in proportion to the review burden.

Problem

Human reviewers lose time when an agent hands off only a diff. The reviewer has to reconstruct the goal, acceptance criteria, verification evidence, risk, and remaining judgment calls before deciding whether the work is mergeable.

Preferred Move

Package the work for review. Include the change summary, acceptance evidence, commands run, known risks, and the artifact that best proves user-facing behavior: tests, logs, screenshots, recordings, benchmarks, specialist review, or deployment notes.

Tradeoff

Do not turn every tiny edit into a release dossier. The packet should be proportional to the risk and should reduce review work, not bury the reviewer in generic ceremony.

Agent Instruction

Before asking for review, produce a compact review packet with the goal, proof, verification, and
remaining risk. Prefer concrete artifacts over confidence language.

Examples

Bad: the handoff asks the reviewer to inspect everything from scratch.

Implemented the checkout fix.

Good: the handoff carries the evidence needed to review quickly.

Fixed checkout retry after payment intent refresh.

Proof:
- Added a regression test that fails on main and passes here.
- Ran the checkout smoke test against Stripe test mode.
- Captured a 45-second recording of the retry path completing.

Risk: did not exercise Apple Pay because the local browser cannot provision that flow.

References

Source Use Note
Symphony README supports Agent work is handed off with proof such as CI and walkthroughs.
Symphony blog supports Human review becomes artifact review instead of session supervision.
OpenAI autonomy loop supports Higher autonomy depends on validation and review evidence.