runpod

name

RunPod

description

GPU pod lifecycle management on RunPod. Create, list, start, stop, and destroy GPU pods for training workloads (TTS voice training, LLM fine-tuning, image generation). Uses runpodctl CLI. <example> Context: User wants to spin up a GPU pod for training user: "create a runpod with an A40 for voice training" </example> <example> Context: User wants to check running pods user: "what pods are running on runpod?" </example> <example> Context: User wants to stop a pod to save money user: "stop the training pod" </example>

RunPod Skill

Manages GPU pod lifecycle on RunPod via runpodctl.

Usage

bash ${CLAUDE_PLUGIN_ROOT}/skills/runpod/runpod.sh <action> [args...]

Actions

Action	Args	Description
list		List all pods (table format)
list-all		List all pods including stopped
get	<pod-id>	Get pod details
create	--name --gpu --image [flags]	Create a new pod
start	<pod-id>	Start a stopped pod
start-spot	<pod-id> --bid <price>	Start as spot instance
stop	<pod-id>	Stop a running pod (preserves data, stops billing)
destroy	<pod-id>	Permanently delete a pod

Create Flags

Flag	Description	Default
--name	Pod name	required
--gpu	GPU type (e.g. "NVIDIA A40")	required
--image	Docker image	required
--gpu-count	Number of GPUs	1
--disk	Container disk in GB	20
--volume	Persistent volume in GB	50
--volume-path	Volume mount point	/runpod
--mem	Min RAM in GB	20
--ports	Exposed ports (e.g. "8888/http")
--env	Environment variables (KEY=VALUE)
--ssh	Enable SSH access	false
--community	Use community cloud	false
--secure	Use secure cloud	false
--cost	Max $/hr ceiling

Examples

# List running pods
bash ${CLAUDE_PLUGIN_ROOT}/skills/runpod/runpod.sh list

# Create a training pod with A40
bash ${CLAUDE_PLUGIN_ROOT}/skills/runpod/runpod.sh create \
  --name "bt-voice-training" \
  --gpu "NVIDIA A40" \
  --image "runpod/pytorch:2.1.0-py3.10-cuda11.8.0-devel-ubuntu22.04" \
  --disk 20 --volume 100 --ssh

# Stop a pod
bash ${CLAUDE_PLUGIN_ROOT}/skills/runpod/runpod.sh stop pod_abc123

# Destroy a pod
bash ${CLAUDE_PLUGIN_ROOT}/skills/runpod/runpod.sh destroy pod_abc123

Notes

• API key is configured in ~/.runpod/config.toml
• Credentials stored in 1Password DEV vault under "runpod"
• Stopping a pod preserves data but stops compute billing
• Destroying is permanent — all data is lost
• Account: [e-mail]