The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from multiple sources, creating a potential surface for indirect prompt injection.\n
Ingestion points: Untrusted content is ingested through moda web search, moda web read, and moda file show for distillation into slide content (references/web.md, references/design-quality.md).\n
Boundary markers: Instructions focus on data preservation but do not implement explicit delimiters or safety instructions to prevent the agent from obeying directives embedded in external content (references/design-quality.md).\n
Capability inventory: The skill has powerful capabilities including canvas modification (moda canvas markup, moda canvas edit), generative media creation (moda media generate-image), and document export (moda export) (references/markup.md, references/edit-code.md, references/export.md).\n
Sanitization: No documented process exists for sanitizing or escaping external content before it is interpolated into agent commands.\n- [DYNAMIC_EXECUTION]: The skill uses a sandboxed JavaScript environment to handle complex batch edits and structural layout changes.\n
Evidence: The moda canvas edit command executes generated JavaScript to mutate canvas nodes, variables, and animations (references/edit-code.md).\n
Mitigation: The execution is restricted to a synchronous sandbox that explicitly blocks dangerous patterns such as eval, require, import, and network-based fetch (references/edit-code.md).\n- [EXTERNAL_DOWNLOADS]: The skill downloads content from the vendor's infrastructure to support multimodal verification and design tasks.\n
Evidence: The skill fetches signed, short-lived preview URLs for brand logos and team templates to local temporary storage (references/brand.md, references/templates.md).

moda-deck