The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses a validation script (scripts/validate-notebook.sh) that executes shell commands to lint, format, and verify notebooks. These commands utilize uvx to run reputable tools like marimo and ruff in isolated environments.
[EXTERNAL_DOWNLOADS]: The data contract (documented in references/data.md) mandates the use of pooch with explicit SHA-256 hash pinning for all external data artifacts, which prevents unauthorized modifications or silent upstream changes.
[REMOTE_CODE_EXECUTION]: The skill executes Python notebooks as part of its core functionality. It mitigates risk by instructing the agent to use the --sandbox flag with the marimo CLI, providing a level of execution isolation.
[DATA_EXFILTRATION]: No patterns of unauthorized data transmission were detected. Network operations are limited to documented data fetching through verified libraries and standard tool updates via uvx.

compose-notebook