AI-assisted product UI often fails before styling: the primary task is unclear, every region becomes an equal-weight card, important interaction states disappear, decorative visuals have no product role, and mobile becomes a compressed desktop layout.
I built Product Design Director, a free and open-source Custom Skill for the Figma agent. Invoke it with /figma-product-design-director.
It asks the agent to work through product context, hierarchy, task-led layout, visual-system decisions, component states, responsive priority, interaction, motion, accessibility review, and a final critique.
The critique format is intentionally practical:
Observed condition → user impact → smallest reasonable revision → recheck the complete flow.
I also ran one controlled paired example using the same starting frame, brief, target frames, and visible runtime configuration, with Skill invocation as the planned variable.
The result was mixed: the Skill-assisted desktop preserved the supplied content more faithfully, while the control surfaced more secondary links in the mobile frame.
This is one observed example, not statistical or causal validation. Agent execution is non-deterministic, accessibility guidance is not conformance evidence, and human design review remains necessary.
I made and maintain the project. It is independently released under the MIT License and is not an official Figma product.
Figma Community:
https://www.figma.com/community/skill/77880/figma-product-design-director
Source and v1.0.1:
https://github.com/Tmmy-inter/figma-product-design-director
Controlled case study:
https://github.com/Tmmy-inter/figma-product-design-director/blob/main/examples/case-study-01-support-operations-workspace.md
The feedback I would value most:
• Which instruction is ambiguous or too restrictive?
• Which product surface or interaction state is missing?
• Where does the Skill conflict with an established design system or authoritative reference?
• Does the critique identify the smallest useful revision?