What is codex-team-mode?

codex-team-mode is a Skill for Codex, the AI coding assistant, that adds a system of four specialized subagents for handling large development, research, analysis, planning, document, data, and content tasks. Instead of one assistant doing everything alone, the main conversation thread can hand off pieces of work to these subagents, while keeping the important decisions and final verification for itself. The four roles are named Explorer, Executor, Complex Executor, and Reviewer. Explorer only reads and gathers information from web sources, documents, code, APIs, logs, and configuration files. Executor handles clear, bounded, low risk work with checks that can be verified automatically. Complex Executor takes on larger implementation tasks once the architecture and safety boundaries are already understood. Reviewer independently checks finished code, plans, reports, or other results from a fresh perspective, without seeing the earlier conversation. Every subagent must be explicitly assigned one of these four roles, and a separate guard rejects any attempt to spawn a subagent without doing so. The tool is described as a routing guide rather than a required pipeline, it can decide to use no subagents at all if delegating them would add more coordination overhead than benefit. The documentation gives detailed rules for when delegation is worthwhile, how to write clear instructions for each subagent, and how to avoid duplicate work or unclear task handoffs. Installation is done by adding the Skill through a package command, then copying five provided configuration files into a personal or per project folder so Codex can find the four agent profiles. Users can customize which underlying model and reasoning effort level each subagent role uses, while the boundaries between read only and mutation capable roles stay fixed. The project is written primarily in Python for its supporting scripts and tests, and is released under the MIT License.