The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/convert_notebook.py utility uses subprocess.run to invoke the marimo CLI tool for converting Jupyter notebooks. The implementation uses a list-based argument structure, which is a standard and secure practice for wrapping CLI tools in Python utilities.
[INDIRECT_PROMPT_INJECTION]: The skill documents and provides templates that ingest untrusted external data, creating a potential surface for indirect injection attacks.
Ingestion points: The assets/data_analysis_template.py template uses mo.ui.file() for user-provided CSV and Parquet files. references/advanced.md describes using mo.sql() for querying dataframes and mo.cli_args() for reading command-line arguments.
Boundary markers: Notebook templates do not currently include explicit delimiters or instructions to ignore embedded prompts in processed data.
Capability inventory: The skill includes scripts for subprocess execution (convert_notebook.py), file system access for persistent caching (references/caching.md), and SQL integration.
Sanitization: While the underlying libraries (like DuckDB for SQL) support parameterization, the provided markdown examples show direct interpolation of variables into queries.
[DYNAMIC_EXECUTION]: Marimo notebooks rely on dynamic execution patterns for reactivity. This includes using mo.lazy() for deferred component loading and lambda functions for UI event handlers like on_click and on_change, as seen in references/ui_components.md and references/advanced.md.
[EXTERNAL_DOWNLOADS]: The documentation references the installation of the marimo and wigglystuff packages via standard package managers (pip, uv). The conversion utility also supports passing GitHub URLs to the marimo convert command, which involves fetching remote notebook content.

marimo-notebooks