菜单
Bootstrap a new curryTrain project by copying the framework Python template into the user's working directory. Use when the user runs /curry-train:init, asks to "start a new training project with curryTrain", "initialize curryTrain", "scaffold a curryTrain project", or wants the framework code copied locally for editing.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Run a short, reproducible benchmark of one optimizer step (forward + backward + optimizer step over N microbatches) using the project's registered runtime. Activate when the user asks to "benchmark a training step", "measure throughput", "time one optimizer step", or "smoke test the runtime". Wraps run_accumulated_step from curry_train.benchmark.
Diagnose a training failure or stall by inspecting recent logs, loss curves, OOM traces, NaN events, and config. Activate when the user asks "why did my training crash", "loss went to NaN", "OOM during step X", "training is not improving", or "help me debug this run". Delegates to the failure-diagnoser agent.
Lightning Fabric integration recipe — minimal 5-line setup that gives DDP / FSDP / mixed precision / mixed-precision while keeping a raw PyTorch training loop. Activate when the user asks "Lightning Fabric", "torchrun", "DDP setup", "FSDP setup", "mixed precision", or wires up the launch script.
Scaffold a new model + dataset + config triple inside an existing curryTrain project. Activate when the user asks to "add a new model to curryTrain", "scaffold a new experiment", "generate model.py / dataset.py / config.yaml for X", or describes an idea they want to start training. Delegates the actual generation to the scaffolder agent.
Compare two training runs and produce a concise markdown diff covering config, key metrics, loss curves, and grad-norm trajectory. Activate when the user asks to "compare run A and run B", "diff two experiments", "did this change actually help", or "is this run better than the previous one". This is both the implementation of the action and the methodology guide for variance-aware comparison.
Define an abort condition before launching a run, so that a clearly broken or clearly under-performing run stops itself instead of consuming the full compute budget. Activate when the user asks "when should I kill a run", "abort condition", "early stop a bad run", "kill criterion", or before any expensive run.
| name | init |
| description | Bootstrap a new curryTrain project by copying the framework Python template into the user's working directory. Use when the user runs /curry-train:init, asks to "start a new training project with curryTrain", "initialize curryTrain", "scaffold a curryTrain project", or wants the framework code copied locally for editing. |
| argument-hint | [target-dir] [--force] |
| allowed-tools | ["Bash","Read","Write","Glob"] |
| disable-model-invocation | true |
Copy the curryTrain Python template from ${CLAUDE_PLUGIN_ROOT}/template/curry_train/ into the user's working directory so they can edit, install, and extend it as part of their own project.
/curry-train:init with an optional target directory.$1 (optional): target directory. Defaults to ./curry_train relative to the current working directory.--force flag: overwrite an existing target directory. Do not assume --force if the user did not pass it.Resolve the target path. If the user supplied an argument, use it. Otherwise default to ${PWD}/curry_train.
Refuse silent overwrite. If the target directory already exists and --force was not supplied, stop and tell the user the directory exists; offer to either remove it or pass --force. Never merge into an existing directory automatically.
Reject self-destructive paths. If the resolved target is inside ${CLAUDE_PLUGIN_ROOT}, abort — this would corrupt the plugin source.
Copy the template.
cp -r "${CLAUDE_PLUGIN_ROOT}/template/curry_train" "<target>"
find "<target>" -name "__pycache__" -type d -exec rm -rf {} +
find "<target>" -name "*.pyc" -delete
Show what was created. Run find <target> -maxdepth 3 -type f | sort and present the file tree so the user immediately sees the layout (Runtime / Primitive / Model / infra / prevalidate / validation / examples).
Print next-step guidance. Tell the user, in this order:
cd into it and pip install -e . if a pyproject.toml exists.new-experiment skill is the natural next step (Stage 1: Skeleton) — they can ask in plain English to "scaffold a new experiment" or "add a model called X".README.md and the stage1-scaffolder skill.pip install, does not git init, does not modify the user's environment, does not download model weights./curry-train:upgrade command may automate it.${CLAUDE_PLUGIN_ROOT} not set or template missing: print the resolved value and abort with a clear message.Keep the post-copy report under 30 lines. Long file trees should be truncated to the top three levels. End with the recommended next command.