runpodctl skill (Runpod): manage Runpod pods and endpoints from the terminal

• List, create, start, stop, restart, update and delete pods

  Create from a template or a Docker image, GPU or CPU. SKILL.md says to set --min-cuda-version 12.8 on every GPU create and to use --wait instead of a poll loop.

  Try: “Create an A100 pod from the PyTorch template and wait until it is ready.”

  Delete · Pods

• Create, update, run, check and delete endpoints

  Create from a template or a Hub listing. serverless run submits a job, status and health check it, and logs reads worker output. A worker minimum of 1 bills continuously, even idle.

  Delete · Serverless

• Find deployable workers and saved pod configs

  hub search finds listings; template search, get, create and delete handle templates. SKILL.md warns that the Hub and pod templates are different catalogs.

  Delete · Hub and templates

• List, create, update and delete storage volumes

  Volumes are tied to one data center. To delete one, remove the pod that uses it first.

  Delete · Network volumes

• Upload and manage your own model files

  model add uploads a local model folder to the Runpod Model Repository; model remove deletes it. --model-reference on serverless create caches a Hugging Face model host-side.

  Delete · Models

• Account, GPU prices and stock, data centers, SSH details

  user shows account info and balance, gpu list shows price and per-data-center stock, and ssh info prints the connection command and key. It does not open a session.

  Runs code · Info and SSH

• send and receive files with a short-lived code

  send prints a one-time code and waits; receive takes that code. Both ends must exit 0.

  Try: “Send ./data to my pod.”

  Runs code · File transfer

• doctor, update, version, completion

  doctor sets up the API key and SSH keys interactively. SKILL.md says to run runpodctl update first, because old builds give confusing errors.

  Runs code · Utilities

• JSON on stdout, coded errors on stderr

  Branch on the code field, such as network_error, rate_limited, not_found or unauthorized. Coded errors need runpodctl 2.8.0 or newer.

  Read · Output and error codes

Watch out for

• Pods and endpoints bill until you stop them

  SKILL.md says a worker minimum of 1 bills continuously even while idle, and a stopped pod still bills for disk and volumes. Delete test resources when done. github.com

• It can delete pods, endpoints, volumes and models

  runpodctl has delete and remove commands for each resource type. Check the id in each command before you approve it. github.com

• The CLI install pipes a script into bash

  SKILL.md offers curl -sSL https://cli.runpod.net | bash, or a Homebrew install as the alternative. Prefer Homebrew or read the script first. github.com

• runpodctl doctor and update change your machine

  doctor writes your API key and sets up SSH keys, and update replaces the binary. SKILL.md tells agents to export RUNPOD_API_KEY instead of running doctor, which is interactive. github.com

• File transfer exposes a code on screen

  send prints a code that lets the receiving side pull the files. Share it only with the pod or machine you mean. github.com

• The plugin route installs all 16 skills and the Runpod MCP config

  The marketplace file lists 16 skills for the runpod plugin. It also bundles .mcp.json, which registers the hosted Runpod MCP server at mcp.runpod.io, and a /runpod:migrate command. The git tree shows no hooks file in the plugin folder. github.com

Security advisories