The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents several interfaces that ingest data from external sources, which constitutes an attack surface for indirect prompt injection. These include file uploads via mo.ui.file, local filesystem browsing via mo.ui.file_browser, and database queries via mo.sql. Additionally, URL query parameters (mo.query_params) and command-line arguments (mo.cli_args) provide routes for external input to influence cell execution. The documentation mitigates this risk by recommending best practices such as escaping user content with html.escape and isolating potentially unsafe scripts using mo.iframe.
[DYNAMIC_EXECUTION]: marimo's core architecture is built around the dynamic, reactive execution of Python code within cells defined by the @app.cell decorator. The skill also details the use of mo.sql for dynamic SQL query generation and execution, which can interact with both local DataFrames and remote databases. These features are integral to the tool's purpose as a reactive programming environment.
[COMMAND_EXECUTION]: The documentation provides a variety of CLI commands for interacting with notebooks, including marimo edit, marimo run, and marimo convert. It also explains how to access raw command-line arguments using sys.argv and parsing them with libraries like argparse or simple-parsing, enabling the creation of configurable scripts and apps.

marimo