github.devotd.
NewCoding agents

yeachan-heo/oh-my-claudecode v5.7.5 — Team-Based Multi-Agent Orchestration

Describes oh-my-claudecode's role-based executors and tmux worker setup for multi-agent orchestration of Claude Code at the team level, the Autopilot workflow and MIT license, and the limitation that the material contains no changelog for v5.7.5.

Repositoryyeachan-heo/oh-my-claudecode ↗

Describes oh-my-claudecode's role-based executors and tmux worker setup for multi-agent orchestration of Claude Code at the team level, the Autopilot workflow and MIT license, and the limitation that the material contains no changelog for v5.7.5.

GitHub description: Teams-first Multi-agent orchestration for Claude Code

stars
★ 40.1K
Forks
3.6K
Language
TypeScript
License
MIT
Latest release
v5.7.6
Read original source ↗View on GitHub ↗

yeachan-heo/oh-my-claudecode is a multi-agent orchestration project that divides multiple AI agents into a team working together in Claude Code. The repository description puts forward a team-first setup. According to GitHub releases, the latest release is v5.7.5.

The README describes this project as multi-agent orchestration for Claude Code. It forms the team from role-specific executors and planning/review stages, and also uses tmux workers (a setup that splits the terminal into sessions for shared use). Installation is split between a Claude Code plugin approach and an npm CLI path, and it supports an Autopilot workflow. It is written in TypeScript and released under the MIT license. The center of this repository is the team configuration that hands out work by role instead of keeping the agent as a single one.

However, the verifiable scope is limited. The star count is 40126 and the repository was created on January 9, 2026. The research record's momentum score is 6.3 out of 10, and no scores for novelty, technology, or quality were given. There is no changelog for v5.7.5 in the material, so it cannot be known what this release changed. There is also no measurement of how much the tmux workers and the Autopilot workflow actually change the working flow. The feature scope and the license are clear, but it is better to approach the project expecting that the concrete behavior of each feature can only be judged within the directly verifiable range.