The Sleek agent skill is one markdown file that turns a coding agent into a mobile design client

One skill, wrapping a paid design API

The Sleek agent skills repository is the official skill for Sleek, described as the AI powered mobile app design tool, and the skill itself is for creating and editing mobile app designs through the REST API.

What the repository contains is small. The top level holds LICENSE, README.md, an assets/ directory with the banner image, and a skills/ directory that contains a single skill, design-mobile-apps, described as being for designing mobile apps, creating screens, and managing Sleek projects with AI. The project description adds that it includes account setup, screenshots and screen HTML.

So this is not a design tool and not an SDK. It is an agent skill, which in the current agent tooling means a directory with a SKILL.md file and whatever supporting material an agent needs in order to drive a particular service correctly. The repository topics point at the intended hosts, with claude-code, codex and cursor listed alongside figma, mobile-app-design, rest-api and ui-design.

The key thing to understand is where the intelligence and the capability actually live. The skill supplies the instructions, the endpoint knowledge and the account setup procedure; the drawing happens on Sleek's servers, and the result comes back as Sleek screens, which is why screenshots and screen HTML are part of the deliverable.

That architecture is what makes this useful and also what makes it a decision with a cost attached, because the tool you are driving is a hosted commercial product with its own pricing, its own plan limits and its own account.

Installing with npx skills add, into your working directory

The installation is an npx invocation, and there are two documented forms.

The first browses and installs interactively:

npx skills add sleekdotdesign/agent-skills

And the second installs the skill directly, by name:

npx skills add sleekdotdesign/agent-skills -s sleek-design-mobile-apps

The skill identifier in that second command is sleek-design-mobile-apps, which does not match the directory name in the repository, design-mobile-apps, so it is worth copying the command rather than reconstructing it.

The installation target is documented, and it is the part most likely to cause a quiet failure. npx skills add installs the skill to .agents/skills/ in your working directory, and if your agent does not discover skills there, you are told to point it at the installed SKILL.md explicitly.

That caveat is a real one for anyone who has used more than one agent. Different agents look in different places for skills, and a skill sitting in a directory your agent never scans is indistinguishable from a skill that is not installed. The workaround is to give the agent the path to the SKILL.md file, which is why the README mentions it explicitly rather than leaving it to the install tool.

It also means the install is per working directory. A project that wants the skill has it in .agents/skills/ inside that project, and a repository you clone elsewhere will not have it until you run the command there.

SLEEK_API_KEY, and a free tier that covers about one run

The requirements section is where the cost of this approach becomes explicit, and it is worth reading before installing anything.

You need a Sleek account. Free accounts can try the API with their one-time trial credits, which the documentation describes as about one design run, and sustained use requires the Pro plan or higher. The pricing given is 69 dollars per month, or 30 dollars per month billed yearly at 360 dollars a year, and that includes 20,000 monthly AI credits.

The API key is stored in the SLEEK_API_KEY environment variable. You create one at sleek.design/agents/setup, which the documentation says handles sign-in, upgrade and key creation in one place, or you manage keys at sleek.design/dashboard/api-keys.

The design of the setup page is worth noting, because it explains the shape of the onboarding the skill itself expects. The account setup is part of the skill's scope, which is why the project description lists account setup alongside screenshots and screen HTML. An agent using this skill can walk a user through creating the account, upgrading the plan and minting the key, then start making designs, all in one flow. That is convenient and it is also the reason the skill is coupled to a commercial funnel.

The credit model has a practical consequence for anyone automating this. One design run on the trial is enough to see whether the output is usable, and not enough to build a workflow on. Twenty thousand monthly AI credits is a ceiling on how many iterations you can afford, and a design loop is iterative by nature, so the number of attempts per screen is the number that decides whether the skill is practical for you.

The skill's surface area is projects, screens and their output

The description of the single skill is short, and it tells you what the skill is meant to cover.

It is for designing mobile apps, creating screens, and managing Sleek projects with AI. Three verbs and one artefact type.

The project description adds the outputs: account setup, screenshots and screen HTML. Those outputs tell you what a caller gets back. A screenshot is a rendered image of the screen, which is what you would paste into a design review or a pull request. Screen HTML is the document behind it, which is what you would inspect when you want to change something specific rather than regenerate the whole thing.

That pairing is the interesting part. An agent that can only produce images is a drawing tool with a chat interface. An agent that can produce a screenshot and the underlying HTML is working on a structured artefact, which means a change request can be expressed as an edit rather than as a reroll. Whether the API in practice supports partial edits is not something the repository documents, so that distinction is worth verifying with a trial run.

What the repository does not contain is equally informative. There is one skill, not a library of them, and there is no code, no SDK package and no test suite in the tree. The assets/ directory holds images for the README. So the quality of this repository is the quality of one instruction document, and reviewing it is reading it.

That also means the format is portable across agents in principle, since a skill is a directory of markdown, but the actual behaviour depends on how faithfully the host agent follows those instructions.

Comparing a REST API skill with a design tool plugin

There are two broad ways to let a coding agent produce mobile design work, and the difference is worth being explicit about.

The route this repository takes is a service API. The agent speaks HTTP to a hosted design service, the service owns the design data, and the output is screens that live in that service's project. The agent is a client. That model has the advantage that the design environment is a real product with a real editor behind it, and the cost that the work leaves your machine and is metered.

The other route is a plugin inside the design tool itself, where the agent drives the tool's own extension surface and the artefact stays a local file that a designer opens and keeps editing by hand. The advantage there is that the file is yours, the tooling is a desktop application your team already pays for or already uses, and the result is not locked to a vendor's project model. The cost is that the integration surface is whatever the tool exposes, which is usually narrower than a REST API, and that the agent's output has to be cleaned up by a human.

The choice is less about quality than about where the design work should live. If designs are going to be iterated in a hosted product and shared through a link, the API route is smoother. If the deliverable is a file a designer will edit further, a plugin or a manual export workflow keeps more of the control in your hands.

What is common to both is the part this repository is silent about, which is how much of the design judgement stays with the agent. Both routes can produce a screen that is technically valid and aesthetically wrong, and neither one removes the need for a designer to look at the result.

MIT for the skill, paid for the service behind it

The licence is MIT, and it applies to the skill, which is the instruction material in the repository. It does not grant anything about the Sleek service, whose use is governed by its own subscription and whose API access is metered by credits.

That distinction is the main commercial fact about this project, and it is a pattern worth recognising across the agent skills ecosystem. The skill is MIT, the tool it drives is a commercial product, and the price of using the skill is the price of the service. A team that adopts it is making a procurement decision as much as a technical one, and the README is unusually direct about the numbers, quoting the monthly and annual Pro pricing and the credit allowance rather than sending you to a pricing page.

The maintenance picture is thin in the way this kind of repository usually is. There are no GitHub releases, the primary language is not identified, and the last push was on 2026-10-05. With a single skill file as the deliverable, there is not much that can change, so a quiet repository is not by itself a warning sign. It does mean there is no version history to consult, and you are relying on whatever the installed SKILL.md says at the moment you install it.

There is also no changelog and no test suite in the tree, so a behavioural change in the service, a renamed endpoint or a changed credit model would reach you through the agent rather than through a release note.

The practical approach for an adopter is to treat the skill as a thin client to a service, not as a tool that stands on its own. That means pinning what you can, reading the installed SKILL.md before your first run, and budgeting the credits for the iteration count your design process actually needs, rather than for the number of screens.

Editorial conclusion

Use the Sleek agent skills if you want a coding agent to produce mobile app screens through the Sleek REST API, and you are prepared to hold a Sleek account with credits on it. Do not adopt it as a free or self-hosted design pipeline, because the free allowance is about one design run and everything beyond that is a paid plan. Before installing, create the API key, check that your agent discovers skills in .agents/skills/ or point it at the installed SKILL.md explicitly, and run one design through the trial credits to see whether the output matches how you work before you plan around it.

Frequently asked questions

Official sources

1. Issues
2. License: MIT
3. Project website
4. README
5. sleekdotdesign/agent-skills on GitHub