From trainhub
把训练任务派发到目标平台（Kaggle kernel / Colab notebook / SSH 远程主机）。适用：在本地准备好代码 + 配置后，推到远端开跑。触发词：提交训练、push kernel、run on kaggle/colab/ssh、派发实验。
npx claudepluginhub shallow-dusty/claude-plugins --plugin trainhubThis skill uses the workspace's default tool permissions.
- 项目已有 `.trainhub.json`（未有 → 先跑 `train-config`）
Creates isolated Git worktrees for feature branches with prioritized directory selection, gitignore safety checks, auto project setup for Node/Python/Rust/Go, and baseline verification.
Executes implementation plans in current session by dispatching fresh subagents per independent task, with two-stage reviews: spec compliance then code quality.
Dispatches parallel agents to independently tackle 2+ tasks like separate test failures or subsystems without shared state or dependencies.
.trainhub.json（未有 → 先跑 train-config）~/.kaggle/kaggle.json；SSH: ssh-config Host；Colab: 浏览器登录）| 参数 | 可选 | 含义 | 示例 |
|---|---|---|---|
platform | 默认读 .trainhub.json | kaggle / colab / ssh | platform=kaggle |
script | 必填 | 训练入口脚本 | kaggle_kernels/L2-ema-p5/l2_ema.py |
run_name | 必填 | 运行唯一命名 | L2-P4-ema-p5-kaggle |
dataset | Kaggle 专属 | 输入数据集 slug | lilliareverie/wall-crack-v2-split |
kernel-metadata.json（用 .trainhub.json.platforms.kaggle 填默认字段 + run_name → id = "{account}/{kernel_prefix}-{run_name}"）kaggle kernels push -p <kernel-dir>rsync 代码到 platforms.ssh.remote_dir/{run_name}/ssh host 'cd remote_dir/run_name && tmux new-session -d -s run_name "conda activate {conda_env} && python script > train.log 2>&1"'半自动：生成好 notebook 复制到 Drive → 指示用户打开 Colab 手动点 Run all（Colab 没有正式 CLI 派发 API）
kaggle kernels list --mine 检查run_name 已存在 → 提示用户决定覆盖/改名打印一条关键信息：下一步怎么 watch
提交成功。下一步：
train-watch run={run_name} platform={platform}
或 /loop 10m train-watch run={run_name}