The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources like user files and existing canvases to generate layouts.
Ingestion points: Untrusted data enters the context through moda file show and moda canvas read calls in the recreation and data preservation workflows described in references/design-quality.md.
Boundary markers: No explicit delimiter strings are used, but instructions command the agent to treat source data as canonical and copy it verbatim, serving as a procedural boundary.
Capability inventory: The skill has capabilities to write to the canvas, delete items, and generate media across all scripts using the moda toolset.
Sanitization: The skill includes strict rules against authoring new content or importing data from memory during the recreation process to prevent instruction following from source data.
[DYNAMIC_EXECUTION]: The skill generates JavaScript for the moda canvas edit tool to perform bulk canvas updates as described in references/edit-code.md.
The scripts are generated based on provided templates to mutate node properties like color, position, and text.
The execution is constrained to a synchronous sandbox that stubs out dangerous APIs like network access, file system access, and process manipulation.

moda-social-instagram