Menú
Manage Google Colab sessions and execute code on remote Colab VMs via the colab-cli.
Instalar con Codex o Claude Copia este prompt, pégalo en Codex, Claude u otro asistente, y deja que revise la página de la skill y la instale por ti.
Basado en la clasificación ocupacional SOC
| name | colab-cli |
| description | Manage Google Colab sessions and execute code on remote Colab VMs via the colab-cli. |
This skill allows an agent to provision, manage, and execute code on Google Colab environments using the colab-cli. It is the preferred way for agents to interact with Colab, supporting both simple scripts and complex ML engineering workflows.
The colab-cli provides a command-line interface that you can invoke via exec or other shell tools.
colab init: Initialize the CLI and authenticate.colab start: Provision a new session.colab exec -f <script.py>: Execute a local script on the remote VM.colab shell "<command>": Run a shell command on the remote VM.colab list: List active sessions.colab stop <session_id>: Stop and delete a session.COLAB_API_KEY or similar is set if required, or assume the environment is already initialized.colab repl). Always use -f for scripts or direct strings for shell commands.colab shell "python -c 'from google.colab import drive; drive.mount(\"/content/drive\")'" if you need Google Drive access.colab-cli may provide paths to extracted artifacts. Check the command output.colab stop when the task is complete to avoid wasting the user's Colab compute units.For MaxEvolve kernel profiling on Google TPUs:
colab start --tpu tpuv6e (or tpuv5e when available in your account)colab shell "pip install -U 'jax[tpu]' -f https://storage.googleapis.com/jax-releases/libtpu_releases.html"profile_script.py and run: colab exec -f profile_script.pyRESULT_LATENCY_MS= / RESULT_TFLOPS= lines for kernel_db_insertcolab stop when doneFor GPU profiling (Hopper/L4/A100), use colab start --gpu and install CUDA-enabled jax[cuda12].
Local configuration and session state are stored in ~/.config/colab-cli/. You can inspect sessions.json there to see persistent session metadata.
MaxEvolve Triton/PyTorch to JAX/Pallas porting with GpuToJax 12-step pipeline, interpret=True validation, MAP-Elites evolution, and Colab/SSH hardware profiling.
Use when you need to generate a synthetic dataset. You can use Afterimage Python package to create, conversational, tool-calling, multi-choice questions, or other types of structured output datasets for SFT, or preference data for DPO.
Schedule deferred or recurring work via the isanagent cron tool—add/remove jobs with correct runtime context fields.
Configure long wall clocks and use execution_run_background + job polling for ML or hours-long runs without blocking the agent loop.
Run numeric or ML code through isanagent execution tools without flooding context—artifacts, incremental runs, sandbox paths.
Turn papers and docs into a concrete training or evaluation recipe (datasets, metrics, hyperparameters, fine-tuning code).