Command Reference
Complete reference for all CROAK CLI commands.
Pipeline Commands
croak init
Initialize a new CROAK project in the current directory. Creates the .croak/ configuration directory, CLAUDE.md project context, and Claude Code skill files.
croak init
croak doctor
Check your environment for required dependencies (Python, GPU, vfrog CLI, etc.) and report any issues.
croak doctor
croak scan
Discover and analyze images in your dataset. Scans for supported image formats and reports statistics.
croak scan # Scan default data/raw/ directory
croak scan ./images # Scan a specific directory
croak validate
Validate data quality and annotations. Checks for minimum image count, class balance, annotation coverage, corrupt images, and bounding box validity.
croak validate
Quality thresholds:
- Minimum 100 images total
- At least 50 images per class
- Maximum 10:1 class imbalance ratio
- 95%+ annotation coverage
croak prepare
Run the complete data preparation pipeline: scan, validate, annotate, and split in one command.
croak prepare
croak annotate
Annotate images using vfrog SSAT (default) or import annotations from external tools.
# vfrog SSAT (recommended)
croak annotate
# Classic import
croak annotate --method classic --format yolo --path ./annotations
croak annotate --method classic --format coco --path ./annotations.json
croak annotate --method classic --format voc --path ./annotations
Options:
| Flag | Description |
|---|---|
--method | vfrog (default) or classic |
--format | Annotation format: yolo, coco, or voc |
--path | Path to annotation files (classic only) |
croak train
Train an object detection model.
croak train # Interactive provider selection
croak train --provider local # Train on local GPU
croak train --provider modal # Train on Modal.com
croak train --provider vfrog # Train on vfrog platform
Options:
| Flag | Description |
|---|---|
--provider | Training provider: local, modal, or vfrog |
--architecture | Model architecture (e.g., yolov8n, yolov11s, rt-detr-l) |
--epochs | Number of training epochs |
--batch-size | Training batch size |
Supported architectures: YOLOv8 (n/s/m/l/x), YOLOv11 (n/s/m/l/x), RT-DETR (l/x)
croak evaluate
Evaluate a trained model's performance. Calculates metrics, performs error analysis, and generates a report.
croak evaluate
Metrics reported:
- mAP@50, mAP@50-95
- Precision, Recall, F1
- Per-class performance breakdown
- Error pattern analysis
croak deploy
Deploy a trained model to production.
croak deploy vfrog # Managed inference API
croak deploy edge # Export to ONNX, TensorRT, CoreML, or TFLite
croak deploy modal # Serverless inference via Modal.com
Status & History
croak status
Show the current pipeline status, including completed stages and next recommended action.
croak status
croak history
Show the iteration history for the current project.
croak history
croak next
Advance to the next SSAT iteration (vfrog workflow only).
croak next
vfrog Integration
croak vfrog setup
Interactive setup for vfrog CLI authentication and context selection. Walks you through login, organisation selection, and project selection.
croak vfrog setup
croak vfrog status
Show the current vfrog CLI authentication and context status.
croak vfrog status
The vfrog CLI uses email/password authentication, which is separate from the VFROG_API_KEY used for inference. See Authentication for API key setup.
Utility Commands
croak upgrade
Upgrade CROAK to the latest version.
croak upgrade
croak help
Show help and available commands.
croak help
croak <command> --help # Help for a specific command
Next Steps
- CROAK Overview — Installation, quick start, and annotation paths
- Claude Code Integration — Use CROAK agents as slash commands