The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions to install the marimo Python package via pip. This is a well-known, legitimate tool for interactive computing and its installation is standard practice. Documentation also mentions fetching remote assets like images, videos, and PDFs from URLs, which is a standard feature for media display.
[INDIRECT_PROMPT_INJECTION]: The skill documents features for ingesting external data that could serve as an attack surface for indirect prompt injection if not handled carefully by the user. As a documentation skill, it neutrally describes these capabilities.
Ingestion points: Documentation in references/inputs.md and references/sql.md covers components that read user-provided files (mo.ui.file), text inputs (mo.ui.text), and remote data sources (mo.sql from URLs).
Boundary markers: The documentation focuses on functionality and does not specify platform-level delimiters, although it provides security best practices like HTML escaping.
Capability inventory: As a notebook environment, the tool inherently supports arbitrary Python code execution, file system operations for caching (references/caching.md), and network operations for fetching media or data.
Sanitization: references/html.md recommends the use of html.escape and mo.iframe to isolate and sanitize potentially untrusted HTML content.
[DYNAMIC_EXECUTION]: The skill describes the notebook's core functionality, which involves reactive execution of Python code. It also mentions mo.persistent_cache in references/caching.md, which uses Python's pickle module for disk-based caching of local objects, a standard feature for this type of development tool.
[COMMAND_EXECUTION]: The SKILL.md file describes legitimate CLI commands for the marimo tool, such as 'marimo edit' and 'marimo run', which are used to manage the notebook environment.

marimo