Run it
Set the project up as Installation describes. Copy the file withbriefs/ and style/ beside it, sign in to Claude Code
and Codex, and run it from that directory:
Terminal
Output, from the offline proof
The file
The brief names the post, the style guide, the grader’s job and the editor’s:briefs/post.md
style/house.md
workflow(): two roles from different model
families, an editor, and two stages:
examples/use-cases/editorial/writer-grader-cap.ts (excerpt)
reviewedBy: 'grade' puts the Codex seat on every draft, and a rejection
runs the draft stage again with the findings in hand. retry: 3 is the
limit: a draft that has not passed after three tries ends the run with the
grader’s last findings, and the editor is never asked. workflow() refuses
the team before any model runs if the writer and the grader share a family.
The publish stage is the editor’s question; a no goes to draft with the
editor’s note as the finding.
Full file
Full file
examples/use-cases/editorial/writer-grader-cap.ts
The team’s shape
What the run did
The proof runs the file against scripted seats and checks the loop the page describes: three writer turns, three grader reads, the superlative and the invented number gone from the draft on disk, and the run stopped at the editor. Obversa recorded the run as one event log, so it shows each draft, each set of findings, and the grader’s pass in order, with the grader’s reasons on the record for the editor to read. The limit is what makes the loop safe to leave running. Without it a writer and a strict grader can go round for as long as the budget lasts. With it, the third refusal is the run’s end and the editor sees the last findings instead of a post. Choose the number for the piece: three suits a short post; a long feature may need more.Next steps
- A writer and a reviewer: the same pair on code, with a real run.
- Feedback loops: the findings, the limit and the other shapes a review can take.
- A person decides: the editor’s question, and how an answer reaches a paused run.