Run Experiment

wanshuiyin/auto-claude-code-research-in-sleep

Automates the entire ML experiment deployment flow across local GPUs, SSH servers, Vast.ai instances, and Modal serverless. It reads your CLAUDE.md to figure out where to run, checks GPU availability, syncs code via rsync or git, spins up screen sessions with the right CUDA device bindings, and optionally integrates W&B logging if you want it. The Vast.ai path is smart about reusing running instances or provisioning new ones with cost analysis, and Modal gets full delegation to a separate skill since it needs no SSH or Docker wrangling. If you're tired of manually sshing into boxes, finding free GPUs, and copying launch commands into screen sessions, this compresses that whole routine into one natural language request. The auto-destroy feature for Vast.ai instances after job completion is a nice touch for cost control.

skills.sh installs: 554
GitHub stars: 18.9K
last updated: Aug 21, 2026

View on GitHub

Install

$ npx -y skills add wanshuiyin/auto-claude-code-research-in-sleep --skill run-experiment --agent claude-code

Installs into .claude/skills of the current project.

Files

Run Experiment

Deploy and run ML experiment: $ARGUMENTS

Workflow

Step 1: Detect Environment

Read the project's CLAUDE.md to determine the experiment environment:

Local GPU (gpu: local): Look for local CUDA/MPS setup info
Remote server (gpu: remote): Look for SSH alias, conda env, code directory
Vast.ai (gpu: vast): Check for vast-instances.json at project root — if a running instance exists, use it. Also check CLAUDE.md for a ## Vast.ai section.
Modal (gpu: modal): Serverless GPU via Modal. No SSH, no Docker, auto scale-to-zero. Delegate to /serverless-modal.

Modal detection: If CLAUDE.md has gpu: modal or a ## Modal section, the entire deployment is handled by /serverless-modal. Jump to Step 4: Deploy (Modal) — Steps 2-3 are not needed (Modal handles code sync and GPU allocation automatically).

SKILL.md on GitHub