Who it is forA Windows developer with a real project and a task too large for one Codex chat, who wants the work split and checked.
The problemOne long Codex session mixes planning, editing, and review, and parallel changes can collide in the main project.
The outcomeA dependency plan, staged worker results, a supervisor review card, and changes written back to the main project only after review.
What you need first
- Windows 10 or 11 (the README says macOS and Linux are not yet supported)
- Codex CLI 0.155 or later, installed and signed in on the same machine
- The portable build `Pet-Office-*-portable.exe` from the GitHub Releases page
Steps
- Download
Pet-Office-*-portable.exefrom the Releases page and run it. - Confirm Codex CLI 0.155 or later is installed and signed in on this machine.
- Hover over the supervisor pet, click the input icon, and turn on 分工 for the task.
- Select the project, the Agent, and the model for each pet, then review and confirm the generated plan.
- Watch the bell task center for approvals, questions, or conflicts until the final review appears on the task card.
Prompt for your agent
Using the Open-Pet-Office repo on Windows, start a 分工 Mission in the desktop app for the project at <project path>. Report the final supervisor review shown on the task card, and verify that a dependency plan was shown and confirmed before any worker ran.
Caveat. The README says the real worker file-write acceptance is still blocked by a Codex Windows sandbox ACL issue in the current candidate build, so expect worker edits may not complete reliably.
Use this in your AI agent
Ask your agent
Using the Deduped MCP, get the setup steps for Gu-kai-lei/Open-Pet-Office and "Split a complex coding task across a supervised agent team"
Add Deduped to Claude Code once
claude mcp add --transport http deduped https://dedupedai.com/mcp
Every Monday: the 5 hottest repos and exactly how to use them. Free, one email a week, unsubscribe anytime.
Get them freeShare: X LinkedIn Hacker News Email Copy link
Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer