Skill v1.0.0

currentLLM-judged scan90/100

oevortex/vtx-coding-agent/google-colab

──Details

PublishedOctober 23, 2026 at 02:35 AM

Content Hashsha256:1308b3261fb925b9...

Git SHA573c71cab6e5

──Files

Files (1 file, 3.5 KB)

SKILL.md3.5 KBactive

SKILL.md · 106 lines · 3.5 KB

version: "1.0.0" name: google-colab description: Manage Google Colab sessions from the terminal using the colab CLI. Run Python code on Colab VMs with CPU/GPU/TPU acceleration, manage sessions, upload/download files, install packages, mount Google Drive, and export session logs. register_cmd: true cmd_info: run code on Google Colab category: cloud

Google Colab CLI

Use colab to manage Google Colab sessions from the terminal. Full reference: https://github.com/googlecolab/google-colab-cli

Install: pip install google-colab-cli

Auth default is ADC. Use --auth oauth2 for browser-based OAuth.

Session Management

bash

colab new                                    # CPU session
colab new --gpu T4                           # GPU (T4, L4, A100, H100)
colab new --tpu v5e1                         # TPU (v5e1, v6e1)
colab new -s my-session --gpu A100           # Named session
colab sessions                               # List active sessions
colab status                                 # All known sessions
colab status -s my-session                   # One session
colab stop -s my-session                     # Terminate
colab url -s my-session                      # Print Colab URL
colab url -s my-session --open               # Open in browser

Code Execution

bash

echo "print('hello')" | colab exec -s my-session          # Pipe code
colab exec -s my-session -f script.py                      # Run local .py
colab exec -s my-session -f notebook.ipynb                  # Run .ipynb
colab exec -s my-session -f script.py --output-image out.png
colab repl -s my-session                                    # Interactive REPL
colab console -s my-session                                 # Raw TTY shell
echo "nvidia-smi" | colab console -s my-session             # Pipe command

One-Shot Script (new + exec + stop)

bash

colab run script.py                                  # Run and destroy
colab run --gpu T4 script.py                         # With GPU
colab run --keep script.py                           # Keep session alive
colab run -s name --gpu A100 script.py -- arg1 arg2  # Named, pass args

File Operations

bash

colab ls -s my-session /content/data
colab upload -s my-session local.csv /content/data.csv
colab download -s my-session /content/result.csv ./result.csv
colab rm -s my-session /content/old.csv
colab edit -s my-session /content/script.py

Packages & Environment

bash

colab install -s my-session -r requirements.txt
colab install -s my-session numpy pandas torch
colab auth -s my-session            # GCP auth (interactive)
colab drivemount -s my-session      # Mount Drive at /content/drive
colab drivemount -s my-session /mnt

History & Logs

bash

colab log -s my-session
colab log -s my-session -n 20
colab log -s my-session -o history.ipynb   # .ipynb, .md, .txt, .jsonl

Utility

bash

colab version
colab update --install
colab help exec

Global Options

Flag	Description
`--auth {oauth2,adc}`	Auth strategy. Default: `adc`
`-c, --client-oauth-config PATH`	OAuth config JSON
`--config PATH`	Session state file
`--logtostderr`	Output to stderr

Notes

exec reads local files and sends them to VM — no upload needed for .py/.ipynb.
repl and console require TTY interactively, but work with piped stdin in scripts.
run is ideal for CI/CD: #!/usr/bin/env colab run --gpu T4
Session state persists in ~/.config/colab-cli/sessions.json.
auth and drivemount require user interaction — cannot be fully automated.

All versions