The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to generate and edit executable Python code in the form of marimo notebooks. This creates a surface for indirect prompt injection if the agent incorporates untrusted data (e.g., file contents, user-provided labels, or external API responses) into the generated code cells.
Ingestion points: User prompts and project files are processed to create notebook cells in SKILL.md and related references.
Boundary markers: The skill does not explicitly instruct the agent to use delimiters or sanitization routines when interpolating external data into notebook code.
Capability inventory: The skill enables full Python execution and shell access via uv run and marimo CLI tools, as documented in SKILL.md and references/DEPLOYMENT.md.
Sanitization: While references/ANYWIDGET.md provides a safety rule against reading sensitive system files (e.g., ~/.ssh, .env), there is no general sanitization logic for data processed within the notebooks.

marimo-notebook