Teach Figma's Agent Your Team's Secret Sauce with Skills

Figma has shipped Skills for its design agent — reusable, plain-English instruction sets that you write once, publish to your team, and trigger any time with a forward slash (/) in chat. If you've ever retyped the same prompt four times in a week, this is the feature you didn't know you were waiting for.

The bigger shift isn't the time saved on prompts, though. It's this: for the first time, you can encode how your team actually works — your brand voice, your crit process, your accessibility checklist — and have the agent apply that knowledge consistently, for everyone, across every project. That changes the AI from a generic assistant into something that actually knows the room.

What a Skill Is (and Isn't)

A Skill is not a plugin, a script, or anything that requires engineering help. It's a set of instructions written in normal language — the kind you'd send in a Slack message — that you teach the Figma agent once. After that, anyone on your team can invoke it by typing / followed by the skill's name in any agent chat.

Skills sit on top of your design system, not instead of it. Figma is clear about the distinction: your design system gives the agent the right components and UI patterns to build with; Skills layer on everything else — brand voice, compliance rules, process rituals, critique frameworks. They're complementary, not competing.

Skills also work inside Figma Make (Figma's app-building environment), so the same reusable workflow follows you whether you're designing on the canvas or generating a working prototype.

Three Skill Types Worth Stealing Right Now

Figma's own teams have been running Skills since the feature launched at Config 2026, and they've shared three categories that translate well to almost any product team.

1. The second-opinion skill

Write a skill that simulates a specific reviewer's feedback style. Figma built one modelled on their CEO's critique approach — feeding the agent examples of his past comments — so designers can pressure-test work before a real review. You could do the same for a demanding stakeholder, a strict accessibility reviewer, or a "first-time user" persona who surfaces friction that anyone close to the product tends to miss.

2. The standards-enforcement skill

Give the agent your style guide and let it take a first pass at catching inconsistencies. Figma's UX Writing team built a skill that checks capitalization, punctuation, and tone against their house standards, so writers can spend their energy on the harder judgment calls instead of hunting for errant title-case.

3. The ritual-automation skill

If your team does something the same way every time, it's a candidate for a skill. Figma describes three that map directly to common design rituals:

• Catch-me-up — summarises recent activity in a file so anyone rejoining after time away doesn't have to dig through comment threads.
• Crit prep — the agent interviews you about your project (persona, scope, audience), then builds a crit page with guided discussion prompts, referencing Nielsen Norman Group best practices.
• Crit recap — takes feedback from a critique, organises it by theme into decisions, action items, and deferred items, and produces a card that can live on the canvas or be pasted into Slack.

That last one alone could save a meaningful chunk of post-meeting admin every sprint.

The MCP Connector Angle

Skills get more powerful when paired with MCP connectors — a protocol (think: a standardised plug) that lets the Figma agent talk to tools your team already uses: Slack, Notion, Asana, and others. That means a crit-recap skill could pull in the actual comments from a Figma file and post the summary to a Slack channel, without you copying anything manually. A crit-prep skill could read a PRD from Notion before it generates the discussion prompts.

This is still relatively new territory, and the depth of each integration will vary depending on what connectors are available and how your team's tools are set up. Worth exploring, but set realistic expectations for the first few weeks.

What to Actually Do This Week

If you want to start small: pick one thing your team re-explains to the AI every single time — a tone-of-voice rule, a component preference, a review checklist — and write it out as a skill. Keep it tight: one clear job, specific enough that the agent can apply it without guessing. Publish it to your team, see how it lands, and iterate.

The ceiling here is genuinely interesting. Encoding institutional knowledge — the stuff that usually lives in a senior designer's head — into something any teammate can invoke is a meaningful step toward making AI a real collaborator rather than just a fast autocomplete. The open question is how well the agent actually honours nuanced instructions over time, especially for complex critique tasks. That's worth watching as your team builds up a skills library.

For now, the floor is low and the upside is real. Start teaching.