
Figma has updated its agent with a new skills system: markdown files that encode recurring design workflows, now discoverable, creatable and shareable directly from Figma Community. Three new features make up this update.
Until now, getting an AI agent to work "properly" in Figma relied heavily on the prompt of the moment. Every designer re-explained, in their own way, how to follow the brand guidelines, structure components, or prepare a clean handoff. The result: inconsistent outcomes, and heavy dependence on whoever was typing the prompt.
Skills change that by turning this know-how into a reusable artifact. You no longer re-explain the context every time: you invoke /skill-name, and the agent applies the already validated protocol. It is the difference between explaining a recipe to someone every morning and handing them a recipe card they follow to the letter.
| Before skills | With skills |
|---|---|
| Prompt rewritten with every use | Workflow encoded once, reused endlessly |
| Result dependent on whoever is prompting | Standardized result across the whole team |
| Know-how confined to one person | Know-how shareable via the Community |
| Compatible only with the Figma agent | Markdown file, portable to other agents |
At Mazette.co, Figma is our daily playground for building interfaces before we industrialize them in Webflow. And one of the real friction points on multi-client projects is precisely consistency: how do you make sure every new prototype follows the same standards for components, naming, and tokens — without re-explaining the same rules with every prompt.
With skills, we can literally lock in our best practices in a file: layer naming structure, design system conventions, a handoff checklist before passing work to dev. It is no longer documentation we hope everyone reads — it is a protocol the agent applies directly.
Another thing we particularly like: the open markdown format. A skill created in Figma can be copied into another agent (Claude, ChatGPT, whatever you use). This avoids the "proprietary silo" effect that we often dread with new AI features — the know-how stays portable, not locked into a single tool.
The library of 50+ community skills is a solid starting point, but it raises a classic question for any open ecosystem: variable quality. A poorly designed skill, published by someone who doesn't fully understand your business context, can introduce bad habits just as easily as a good skill fixes them. As with Figma's historic plugins, you will need to develop a habit of vetting and testing before adopting a third-party skill in production.
Good idea, clearly. It is the logical continuation of everything Figma has been building around its agent over the past few months — giving context, memory, and repeatability to a tool that, without it, remains a good assistant but a poor colleague. Skills turn the agent into an actual team member who knows your conventions, rather than an intern you have to re-brief every Monday.
For an agency juggling several brands and several client design systems in parallel, this is a real gain in reliability and speed. We are keeping a close eye on how this library evolves, but the intention behind it — capitalizing on know-how instead of losing it with every prompt — is exactly the right direction.
Code