run-experiment

Deploy and run ML experiments on local or remote GPU servers.

1|Updated Mar 24, 2026

Search Tags:#automation #orchestration #gpu #deploy #ssh #wandb #ml-experiments

Ready to run

npx skills add https://github.com/kitcaf/skills --skill run-experiment-kitcaf

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: run-experiment
Source: https://github.com/kitcaf/skills/tree/main/skills/skills-codex/skills/run-experiment
Command: npx skills add https://github.com/kitcaf/skills --skill run-experiment-kitcaf

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Deploy and run ML experiments on local or remote GPU servers, handling environment detection, pre-flight checks, and remote/local execution.

Core Features & Use Cases

Environment detection based on AGENTS.md to determine whether to run locally or on a remote GPU server
Pre-flight GPU checks to verify availability before launching jobs
Code synchronization and deployment workflow with optional WandB integration
Remote execution via SSH with screen sessions and GPU binding for long-running experiments
Local execution support with CUDA/MPS setups and conda environments
Optional Feishu/W&B style notifications after successful launches

Quick Start

Deploy and run an ML experiment by specifying the target script and its arguments.

Frequently Asked Questions about run-experiment

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I run ML experiments on a remote GPU server via SSH?▼

To run ML experiments on a remote GPU server via SSH, this Skill automates code synchronization, pre-flight GPU checks, and execution using screen-backed sessions for reliable long-running deployments.

Can I deploy ML experiments locally with CUDA or MPS setups?▼

Yes, you can deploy ML experiments locally with CUDA or MPS setups. The Skill detects the environment based on AGENTS.md configurations and supports execution within local conda environments.

Does this workflow support experiment tracking with wandb?▼

Yes, the deployment workflow supports optional experiment tracking with wandb. It integrates wandb during code synchronization and execution to monitor ML experiments.

What is the best way to ensure long-running ML training jobs do not disconnect?▼

The best way to ensure long-running ML training jobs do not disconnect is using screen-backed sessions. This Skill leverages screen for remote SSH execution to maintain reliable, persistent GPU-bound training.

Do I need to manually verify GPU availability before launching my training script?▼

No, you do not need to manually verify GPU availability. The Skill performs pre-flight GPU checks automatically before launching jobs to ensure resources are ready for your ML experiments.

How does environment detection work for ML deployment targets?▼

Environment detection for ML deployment targets works by reading configurations specified in AGENTS.md. This determines whether your experiment runs locally or deploys to a remote GPU server.