Menu
Interpret Jupyter execution sessions in isanagent—attachments, kernel cwd, tokens; contrast with local provider behavior.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
MaxEvolve Triton/PyTorch to JAX/Pallas porting with GpuToJax 12-step pipeline, interpret=True validation, MAP-Elites evolution, and Colab/SSH hardware profiling.
Manage Google Colab sessions and execute code on remote Colab VMs via the colab-cli.
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.
| name | jupyter-heavy-output |
| description | Interpret Jupyter execution sessions in isanagent—attachments, kernel cwd, tokens; contrast with local provider behavior. |
| requires | {"bins":[],"env":[]} |
| always | false |
Apply this when execution_env_info (or the session you opened) shows the Jupyter provider—i.e. the host configured default_provider = "jupyter" and valid [harness.execution.jupyter] settings.
execution_session_create returns session_capabilities (JSON) for that session. Jupyter sessions use a remote kernel with persistent variables across execution_run calls until the session is closed or the kernel dies.execution_run may return attachments: sandbox-relative paths for materialized image/png, image/jpeg, large text/csv, large application/json, etc. Short human-readable lines may appear in stdout; do not assume full binary or huge JSON appears in the tool string—open paths via execution_artifact_list + file tools, and describe outcomes in prose.execution_run only supports cwd_mode: session_default in practice—the process cwd is on the Jupyter server, not your agent sandbox. If the user needs a server-side directory change, encode it in code (e.g. os.chdir / %cd) explicitly; do not confuse server paths with sandbox_relative cwd for the local provider.| Concern | Local provider | Jupyter provider |
|---|---|---|
| Where code runs | Child / REPL worker on agent host, cwd under sandbox | Remote kernel on Jupyter server |
| Packages | Same as agent host python | Whatever is installed on the server |
| Big plot/table | Prefer write under .execution_artifacts/... or attachments pattern | Often attachments from display_data |
| Interrupt | execution_cancel + harness behavior | Uses server interrupt when supported |
Check execution_session_create metadata (e.g. extensions.local_python_repl on local) when you need to remember whether local Python is REPL-style or subprocess—do not guess from memory.
Prefer the JUPYTER_TOKEN environment variable on the agent process over pasting tokens into replies. Never echo tokens into chat or logs.