More agents. Clearer boundaries.
A reusable workflow for assigning work, reviewing changes, and keeping human decisions visible.
The problem.
Different coding agents working in the same checkout were overwriting one another’s changes. I needed a way to keep each agent’s work separate, make reviews and approvals explicit, and preserve context as tasks moved between sessions.
What I built.
I built Agent Coordinator around separate workspaces, scoped tasks, written handoffs, explicit approvals, and independent review. It supports Claude Code and Codex workflows, including review across model providers.
The decisions that matter.
Separate the work
Give each agent a scoped task and an isolated workspace, keeping concurrent changes separate until they are deliberately integrated.
Ask for another view
Make independent review part of the workflow, including reviews from a different provider.
Keep decisions durable
Record approvals and handoffs so they remain inspectable when work moves between sessions.
The result.
A reusable public plugin that makes task, review, and approval boundaries easier to follow. Several agents can work at once, and the human’s approvals stay written down instead of assumed.
Built with.
- Markdown
- Shell
- Python
- Claude Code & Codex