Valohai Agent Skills

A collection of Agent Skills for migrating ML codebases to Valohai, the MLOps platform.

These skills give AI coding agents (Claude Code, GitHub Copilot, Cursor, etc.) the knowledge to help users migrate their existing ML projects to Valohai with minimal code changes.

Skills

Skill	Description
valohai-migrate-parameters	Convert hardcoded values to Valohai-managed parameters with argparse + YAML
valohai-migrate-metrics	Add JSON-based metrics tracking for experiment comparison and visualization
valohai-migrate-data	Migrate data loading/saving to Valohai's input/output system
valohai-yaml-step	Create valohai.yaml step definitions with Docker images, commands, and configuration
valohai-project-run	Set up Valohai projects and run executions/pipelines via the CLI
valohai-design-pipelines	Analyze ML projects and design multi-step Valohai pipelines

Installation

npx skills add valohai/valohai-skills --all

Migration Workflow

The recommended order for migrating an existing ML project:

1. Analyze the project structure and identify scripts, dependencies, and data flow
2. Create steps (valohai-yaml-step) - Define Docker image, commands, and basic YAML
3. Add parameters (valohai-migrate-parameters) - Make hyperparameters configurable
4. Add metrics (valohai-migrate-metrics) - Enable experiment tracking with JSON printing
5. Migrate data (valohai-migrate-data) - Set up inputs from cloud storage and outputs to /valohai/outputs/
6. Set up project (valohai-project-run) - Create project, link directory, run first execution
7. Design pipelines (valohai-design-pipelines) - Connect steps into automated workflows

Philosophy

Valohai follows a YAML-over-SDK approach: your ML code should not be entangled with the platform. Configuration lives in valohai.yaml, and your code remains portable and framework-agnostic.

License

Apache-2.0