The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process data from external research, web browsing, and delegated worker agents, which could potentially be manipulated to include malicious instructions.
Ingestion points: Research findings and worker outputs described in references/delegation.md, external product implementations in references/style-grounding.md, and asset descriptions in references/images.md.
Boundary markers: The skill defines a 'delegation gate' in references/delegation.md requiring workers to adhere to specific, compact output contracts to limit their influence.
Capability inventory: The skill possesses powerful document mutation capabilities via the apply_canvas tool and task management via manage_design_task.
Sanitization: The instructions prioritize human-in-the-loop verification where the main agent must inspect rendered pixels (get_screenshot) and native document facts (get_structure) rather than blindly trusting the input data.
[EXTERNAL_DOWNLOADS]: The skill facilitates the retrieval of visual assets such as images and SVGs from arbitrary remote URLs.
Evidence: references/images.md and references/icons.md describe the process for using imageUrl and external SVG definitions to populate Figma designs.

design-in-figma