Part A: a model reviews your partner's work

You get a second opinion from a model on your partner’s work. Model review is advisory, and your own decision stays the only acceptance.

Principles in play: Review against sources, Accept independently.TRACEReview against sources · Accept independently

Needs: the hour behind you; the second model also needs your OpenRouter key.

What you do

1. Say to your agent

   Review my partner’s candidate with the review skill, trace-review, against the task in the workshop guide and its source, and show me the row.

   If the reviewer cannot be started, your agent says which step failed and stops.

2. Read the review your agent copies into your sandbox as review-of-A.md or review-of-B.md. Each finding is labelled by whether its quote is in the candidate word for word.

3. Open ai/records/acceptances.jsonl. The new row says model review; your own row is still the only decision.

4. If your OpenRouter key is in .env after OPENROUTER_API_KEY=:

   Say to your agent

   Prepare a packet for the second model on the same candidate, and show me what would be sent.

   You choose the model; the workshop suggests glm5.3 or kimi k3.

5. Say to your agent

   I have read the draft. Finalize it in my name, Your Name, then commit and push.

Ask your agent

• Which of the findings quote the candidate word for word, and which do not?
• Whose name is on the review row, whose is on the decision row, and which is an acceptance?
• Is the review row marked self-review, and why?

What you can say now

A model reviewed my partner’s work against a criterion and its sources: that is R, review against sources. Its row says model review, which is advisory. My own row is still the only decision, made by someone independent of production: that is A, accept independently, left with the person.