<p align="right"> <strong>English</strong> · <a href="./README.zh-CN.md">简体中文</a> </p>

<p align="center"> <img src="./assets/readme/agent-map.svg" width="100%" alt="Four character-driven Team Mode agents divide exploration, bounded execution, complex execution, and independent review while the main thread leads and verifies the work."> </p>

`team-mode` is a Codex Skill for coordinating four working agents across substantial development, research, analysis, planning, document, data, and content tasks. The main thread keeps unresolved decisions and verifies the final result; subagents take on work that benefits from focused context, lower cost, safe parallelism, or independent judgment. A separate low-cost `default` guard rejects any spawn that omits `agent_type`.

It is a routing guide, not a mandatory pipeline.

## The team

- **Explorer（探索者）· Luna Medium · read-only** — gathers evidence across current web sources, documents, data, code, APIs, logs, and configuration. - **Executor（执行者）· Luna High · workspace-write** — completes clear, bounded, low-risk work with deterministic checks. - **Complex Executor（复杂执行者）· Terra High · workspace-write** — completes substantial but bounded implementation after architecture, acceptance checks, and safety boundaries are clear. - **Reviewer（复审者）· Sol High · read-only** — independently checks stable code, reports, plans, analyses, data, and other artifacts from fresh context.

Every spawn must explicitly pass one of those four names through `agent_type`. `task_name` is only a label, and `default` is never a working role.

Luna keeps exploration economical and gives bounded execution a higher reasoning margin. Terra handles most conventional complex implementation while the main thread retains architecture decisions and final acceptance. Sol is reserved for focused independent review of concrete consequential risks.

The TOML sandbox is a profile default, not a guaranteed isolation boundary: a live parent permission override c