One agent writes, another reviews

Many agents, one playbookDeveloper who uses Codex as a second pair of eyes

We asked Codex to review Claude's change. It flagged the cents-based prices as 'probably a bug', which was a deliberate decision.

How do I stop Codex from flagging deliberate choices as bugs when it reviews Claude's code?

Record your team's decisions with ArcRoom, and the reviewing agent gets them in recall along with what the writing agent changed and why. Codex then skips questions that are already decided, like prices kept as integer cents, and focuses on real issues. Decisions have to be recorded to be respected.

Also asked as: My second AI reviewer keeps flagging things we decided on purpose. How do I stop it? · How do I give a reviewing AI agent the reasons behind past decisions? · Can Codex review Claude Code's changes with the team's decisions in mind?

The work

The reviewer gets the team's decisions in recall.

ArcRoom keeps
  • Decision
  • Recall
The next agent gets

[codex review] Decision on file: prices are integer cents (why: rounding). Not flagging.

Step by step

  1. The reviewer gets the team's decisions in recall.
  2. It sees what the writing agent changed and why it saved those steps.
  3. Review comments skip decided questions and focus on real issues.

What the agent sees

[codex review] Decision on file: prices are integer cents (why: rounding). Not flagging.

Compared with today's tools

Decisions pasted into the review prompt

Paste the relevant decisions into the review prompt, or list them in AGENTS.md for Codex.

Setup
10 min
Ongoing
~3 min per review to find and paste

ArcRoom

Every agent reads the team playbook before it works and writes to it when it finishes. Nothing to update by hand.

Setup
2 min per agent (npx arcroom init --agent …)
Ongoing
None

Close if the decision list is current; ArcRoom picks the matching decision for each review without anyone pasting.

Where it stops

Decisions have to be recorded to be respected.

Try it in your project

Works with the agents your team already uses, and any agent over MCP or the HTTP API.

More in many agents, one playbook