verify-training-pipeline

npx skills add majiayu000/claude-skill-registry --skill verify-training-pipeline-kungfuai-cvlization --agent claude-code

Same command for any agent — swap --agent for codex, cursor, copilot.

# Verify Training Pipeline Systematically verify that a CVlization training example is complete, properly structured, and functional. ## When to Use - Validating a new or modified training example - Debugging training pipeline issues - Ensuring example completeness before commits - Verifying example works after CVlization updates ## Important Context **Shared GPU Environment**: This machine may be used by multiple users simultaneously. Before running GPU-intensive training: 1. Check GPU memory availability with `nvidia-smi` 2. Wait for sufficient VRAM and low GPU utilization if needed 3. Consider stopping other processes if you have permission 4. If CUDA OOM errors occur, wait and retry when GPU is less busy ## Verification Checklist ### 1. Structure Verification Check that the example directory contains all required files: ```bash # Navigate to example directory cd examples/<capability>/<task>/<framework>/ # Expected structure: # . # ├── example.yaml # Required: CVL metadata # ├── Dockerfile # Required: Container definition # ├── build.sh # Required: Build script # ├── train.sh # Required: Training script # ├── train.py # Required: Training code # ├── README.md # Recommended: Docu

Navigate to example directory
cd examples/<capability>/<task>/<framework>/
Expected structure:
Option 1: Build using script directly
Option 2: Build using CVL CLI (recommended)
cvl run <example-name> build
Verify image was created
docker images | grep <example-name>
Option 1: Run training using script directly
Option 2: Run training using CVL CLI (recommended)