The Agent Skills Directory

[SAFE]: The skill serves as a reference guide for the marimo library. All documented CLI commands and Python APIs are standard features for the platform's intended use in data science and application development.
[INDIRECT_PROMPT_INJECTION]: The library exposes numerous ingestion points for external data, such as UI text inputs, file uploads, and URL query parameters. This represents an attack surface for indirect prompt injection; however, the documentation appropriately includes security best practices, such as content sanitization via html.escape and isolation using mo.iframe.
Ingestion points: mo.ui.text, mo.ui.file, mo.ui.file_browser (found in references/inputs.md), mo.query_params (found in references/query-params.md), mo.app_meta().request (found in references/app.md), and mo.watch.file (found in references/watch.md).
Boundary markers: The documentation suggests using markdown formatting and explicit HTML callouts to distinguish content.
Capability inventory: Python code execution, filesystem access via mo.ui.file_browser, network requests, and database querying via mo.sql.
Sanitization: Recommendations for using html.escape and mo.iframe are provided in references/html.md to handle potentially untrusted content safely.
[DYNAMIC_EXECUTION]: The skill describes methods for programmatic cell execution (cell.run()) and persistent caching using pickle. While pickle deserialization can be risky if data is tampered with, it is documented here as a local performance optimization tool, which is a standard practice. The library also supports json as a safer alternative for caching.

marimo