Can I control which models the teammates use?

Yes, you can specify the model for the entire team or individual teammates (e.g., using Sonnet for all) via natural language prompts when starting the team.

How do agent teams differ from standard subagents?

While subagents report results back to a single caller, Agent Teams are fully independent, can message each other directly, and coordinate via a shared task list for complex, long-running work.

Do I need special software to run agent teams?

To view teammates in split panes, you need tmux or iTerm2. However, the skill also supports an 'in-process' mode that works in any standard terminal by cycling through agents.

Can teammates work on the same file?

It is recommended that each teammate owns different files or modules to avoid merge conflicts, though they can use the mailbox system to coordinate on shared dependencies.

How many agents should I have in a team?

Best practices suggest starting with 3-5 teammates. This balances the benefits of parallelism with the coordination overhead and token costs associated with multiple instances.

Claude Code Agent Teams

Name: Claude Code Agent Teams
Author: dwmkerr

bydwmkerr

•

协作工具

Orchestrates multiple Claude Code instances to work in parallel with shared task management and inter-agent communication.

Claude Code Agent Teams allows you to scale engineering productivity by spawning multiple independent Claude sessions that coordinate as a unified team. Unlike simple subagents, these teammates can message each other directly, claim items from a shared task list, and work on different layers of a project—such as frontend, backend, and testing—simultaneously. This skill is perfect for complex tasks requiring diverse perspectives, such as architectural research, multi-module refactoring, or parallel debugging of competing hypotheses.

主要功能

01Flexible display modes including in-process cycling and tmux split panes.

02Multi-agent orchestration with shared task lists and self-coordination.

03Inter-agent messaging for collaborative problem-solving and peer review.

04Context-aware task claiming to prevent file conflicts and redundant work.

0512 GitHub stars

06Quality gate integration via TaskCompleted and TeammateIdle hooks.

使用场景

01Conducting multi-angle research and technical reviews simultaneously.

02Executing high-speed debugging by testing multiple hypotheses in parallel.

03Parallelizing development across frontend, backend, and infrastructure layers.

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add dwmkerr/claude-toolkit claude-code-agent-teams

For use in Claude.ai and ChatGPT

主要功能

01Flexible display modes including in-process cycling and tmux split panes.

02Multi-agent orchestration with shared task lists and self-coordination.

03Inter-agent messaging for collaborative problem-solving and peer review.

04Context-aware task claiming to prevent file conflicts and redundant work.

0512 GitHub stars

06Quality gate integration via TaskCompleted and TeammateIdle hooks.

使用场景

01Conducting multi-angle research and technical reviews simultaneously.

02Executing high-speed debugging by testing multiple hypotheses in parallel.

03Parallelizing development across frontend, backend, and infrastructure layers.

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add dwmkerr/claude-toolkit claude-code-agent-teams

For use in Claude.ai and ChatGPT