A skill only repeats a method a designer already decided. The designers who write the skills end up defining how the whole team works with the agent.
Figma agent custom skills: the short answer
Figma agent custom skills are reusable instructions you write once and call with a slash command in any Figma agent chat. When the Figma agent became generally available on 6 October 2026, Figma's release notes listed custom skills among the GA features and said they are available on the Starter plan, and the Figma help center guide to custom skills says they are available on all plans and work in both Figma Design and Figma Make.
The short version of how to use them as a designer: turn the prompts you keep retyping (crit prep, UX copy checks, accessibility passes, handoff notes) into a skill with a clear trigger, numbered steps and a definition of done. Keep each skill narrow. Publish it to your team only after it has worked on three real files. This playbook walks through that process, where skills break, and what they mean for the job.
What Figma actually shipped
Skills first appeared during the agent's open beta. Figma's product team introduced them in a July 2026 post, Got skills? Make the Figma agent a better collaborator, with examples such as a stakeholder feedback simulator, a UX writing standards check, a new-user perspective review, crit prep and recap, and a skill that turns review feedback into tasks. The 6 October GA release opened them up more widely, and the agent now consumes AI credits.
The mechanics, according to the help center:
Three ways to create a skill. Ask the agent to write one (for example from a conversation you just had, or by merging two skills), upload an existing file, or write it manually with a name, a description and Markdown instructions.
The name is the command. Whatever you call the skill becomes its slash command.
Uploads are a single Markdown file that follows the Agent Skills specification. Folders such as scripts, references and assets are not supported.
Sharing. Publish privately to a team or organization, or publicly to Figma Community with a thumbnail, description and support contact. Updates go out with Publish changes.
Management. Edit, export, disable or delete from Add context, then Skills, then Manage skills. Deleting a published skill removes it for everyone, so export first.
One known limit. If a prompt mentions several skills, only the first one used runs.
Skills sit next to the library guidelines covered in the Figma agent guide for designers. Figma's own framing is that the two are complementary: guidelines tell the agent how your design system works, while skills carry the rest of the team's know-how, such as voice, review methods and process.
Hype vs reality
The hype says skills turn the agent into a junior designer who knows your team. That overstates it. A skill is a saved prompt with a trigger. It does not learn from use, it does not remember what went wrong last week, and it cannot do anything the agent itself cannot do.
What holds up is consistency. The most expensive part of working with any agent is re-explaining context, and most of that context is boring and repeatable: how your team names layers, which copy rules apply, what a crit page should contain. A skill captures that once. When ten designers run the same /ux-copy-check, they get the same checklist applied the same way. That is a real gain, and it is the same pattern covered in Framer skills for designers: written instructions become a team asset.
Verdict: worth adopting now for review and hygiene tasks. Not worth building a large library before you know which skills people actually call.
Prompting vs custom skills vs library guidelines
Question | One-off prompt | Custom skill | Library guidelines |
|---|---|---|---|
What it holds | A request for this chat | A repeatable task or method | Design system rules |
How it runs | You type it each time | Slash command in any file | The agent reads it on every prompt |
Best for | Exploration, unusual tasks | Crits, copy checks, handoff, audits | Component usage, spacing, tokens |
Shared how | Copy and paste | Publish to team, org or Community | Attached to the library |
Main risk | Inconsistent results | Vague triggers, bloated steps | Rules nobody updates |
Use all three. Guidelines for what to build with, skills for how your team works, plain prompts for everything new.
How to use Figma agent custom skills: a step-by-step playbook
1. Mine your chat history first. Look at the last two weeks of agent chats and your own notes. Any prompt you typed more than three times is a skill candidate. Common ones for product designers: prepare a crit page, check UX copy against a style guide, review a flow as a first-time user, list missing empty, loading and error states, write handoff notes for engineering.
2. Let the agent draft, then rewrite it yourself. The help center explicitly supports turning a conversation into a skill. Do that after a chat that went well, then edit the result. Agent-written skills tend to be wordy and generic. Cut anything that does not change the output.
3. Use the Trigger, Steps, Proof template. Every skill needs three parts:
Trigger: one sentence saying when to use it. Figma's guidance is to phrase this as a requirement rather than a restriction so it fires reliably. Write "Use this skill when reviewing UX copy on a selected frame", not "Do not use this outside copy reviews".
Steps: a numbered list the agent follows in order, with the exact checks. "Sentence case for buttons, no trailing periods on single-line labels, error messages say what happened and what to do next."
Proof: what done looks like. "Return a table of issues with the layer name, the current text and a suggested fix. Do not edit the canvas."
The Proof line matters most. It stops the agent from quietly changing your file when you only asked for a review.
4. Name it so it does not collide. The name becomes the slash command. If you will publish, Figma recommends a specific name, for example adding your team or company. /acme-copy-check beats /check.
5. Test on three real files before publishing. Run it on a clean file, a messy file and an edge case. Keep a list of every miss and fix the instruction that caused it. Only then publish privately to your team.
6. Run one skill per prompt. Because only the first skill in a prompt runs, chaining /crit-prep and /copy-check in one message will not do what you expect. Run them in sequence.
7. Review and prune monthly. Disable skills nobody calls. Export anything before deleting, since deleting a published skill removes it for the whole team.
Where skills fail
Skills fail when the trigger is vague, when one skill tries to do five jobs, and when the steps depend on judgement the agent does not have. A skill that says "make the design better" is just a slower prompt. A skill that encodes a stakeholder's taste can also mislead a junior designer into designing for the simulator instead of the user. Treat simulated feedback as a rehearsal, never as approval.
Also check cost. Since GA the agent uses AI credits, so a skill that runs a full-file audit on every small change can burn through a team's allowance. Scope skills to a selection where you can.
Where design work is going
Over the next one to three years, expect a team's way of working to live in files an agent can read: guidelines for the system, skills for the process. This is the same direction explained in design systems for AI. Writing a good skill is closer to writing a good design spec than to prompt tricks: clear inputs, explicit rules, a testable result. Designers who are good at that will set how their whole team works with agents.
The fear: will Figma agent skills replace designers?
No, and the reason is in how skills are built. Someone has to know what a good crit looks like, which copy rules matter, and which states engineers always forget. The skill only repeats what a designer already decided. If you are worried about falling behind, the honest risk is not that a skill replaces you. It is that a teammate who writes the skills ends up defining the process everyone else follows.
Moves for this week: pick the one review you do most often, write it as a Trigger, Steps, Proof skill, test it on three files and share it with one teammate.
If you want help setting up an agent-ready design system and the product work around it, DIGCY offers UI/UX design services, product design services and AI product development. DIGCY is a Dribbble Selected Agency and a UI/UX agency in Casablanca working with clients globally.
FAQ
What are Figma agent custom skills?
Figma agent custom skills are reusable Markdown instructions you invoke with a slash command in a Figma agent chat. They package a repeatable task, such as a UX copy check or crit prep, so you do not have to rewrite the prompt each time.
Are Figma agent custom skills available on the free plan?
Figma's help center says custom skills are available on all plans, and the 6 October 2026 release notes state they are available on Starter. The agent itself uses AI credits since GA, so usage depends on your credit allowance.
How do I create a custom skill in Figma?
Click the prompt box, choose Skills, then Add skill. You can start from scratch with a name, description and Markdown instructions, upload a single Markdown file that follows the Agent Skills specification, or ask the agent to write a skill from a conversation.
What is the difference between Figma skills and library guidelines?
Library guidelines describe your design system and the agent reads them on every prompt. Skills describe a task or process and only run when you call them with a slash command.
Will Figma agent skills replace UI designers?
No. A skill repeats a method a designer already defined. It cannot decide what a product should do, judge whether a flow helps users or write good rules on its own.
Are Figma agent skills overhyped?
Partly. They are saved instructions, not a teammate that learns. The real value is consistent reviews and less re-explaining, which is worth having for any team using the agent daily.
Let’s keep in touch.
Discover more about high-performance web design. Follow us on Twitter and Instagram.
