Skip to content

Teach your agent creative craft

14 skills that answer the questions a tool list leaves open — which of 300+ models for this job, what the prompt has to name, why the result is not landing.

View on GitHub

Open source and MIT-licensed. Works with Claude Code, Cursor, Codex, VS Code, and any agent that reads a SKILL.md.

Two steps, once per project

The skills carry the procedure. The MCP server is what carries it out — they are a pair, not alternatives.

  1. 1

    Add the skills

    npx skills add layerai/skills --skill "*"

    Pulls layerai/skills into your project. Keep the --skill "*": without it the command opens a picker with nothing preselected, and pressing enter installs nothing.

  2. 2

    Connect the MCP server

    claude mcp add --transport http Layer https://mcp.app.layer.ai/mcp

    OAuth with your Layer account, no API key. Other clients and tokens.

In Claude Code, one step does both

/plugin marketplace add layerai/skills

Adds the pack as a plugin marketplace and wires up the MCP server at the same time.

What your agent learns

One document per craft. Your agent loads the one that matches the work, so the asset that comes back is usable rather than merely generated.

Getting started

Connect to Layer and learn the generation loop every other skill builds on: model discovery, the spend gate, uploads, and polling.

layer

The core loop

Choosing a base model, pricing a run in Creative Units before it spends them, uploading references, polling, and recovering when a call comes back wrong.

Read the skill

Direction

Agree a look before generating at volume: moodboards, storyboards, colour, era, and framing.

layer-art-direction

Art direction

Moodboards, storyboards with one beat a panel, a pinned palette, and naming a style or era precisely enough to generate against.

Read the skill

Images

Generate stills and change the ones you already have: prompting, sizing, references, masked edits, reframing, upscaling, and layers.

layer-image

Still images

Concept art, key art, icons and marketing stills — what a prompt has to name, and how to make a set of images share one look.

Read the skill
layer-image-editing

Editing an image

Masked edits, outpainting, reframing, relighting, restyling, background removal, splitting into layers, and upscaling.

Read the skill

Game art

Art that has to sit in a system: characters and casts, environments and parallax, game UI, pixel sprites and sheets, and seamless textures.

layer-game-assets

Characters and UI

A cast that stays on-model across a dozen images, environments and parallax layers, and icons that still read at 64px.

Read the skill
layer-pixel-art

Pixel art

Sprites on a real grid rather than blurred ones, walk cycles, and packing frames into a sheet or cutting one back up.

Read the skill
layer-textures

Tiling textures

Materials that repeat without a visible seam, without an obvious motif across the surface, and without lighting baked in to fight the engine.

Read the skill

Video and audio

Motion and sound: animating stills, camera and loops, sound effects, music, speech, and cutting shots into one finished video.

layer-video

Video

Animating a still, camera motion, extending a clip, loops, lip sync, and planning a multi-shot ad, trailer, or cutscene.

Read the skill
layer-audio

Sound and speech

Sound effects without unwanted ambience, music with structure and length control, and choosing a voice and delivery for narration.

Read the skill
layer-video-timeline

Cutting a video

Sequencing generated shots, transitions, captions, a logo or CTA, and rendering the whole thing to one deliverable file.

Read the skill

3D

Meshes for a game engine: generation from text or image, remeshing, retexturing, rigging, and animation.

layer-3d

3D meshes

Text or image to mesh, quad against triangle output, a polygon budget, PBR textures, rigging, and animating what comes back.

Read the skill

Consistency and pipelines

Hold one look across a whole project with trained reference sets, and run the Blueprint workflows a studio has already built.

layer-reference-sets

Trained styles

When to train a LoRA rather than attach a reference, what to curate into it, what weight to run it at, and how to tell when it is being ignored.

Read the skill
layer-workflows

Blueprint workflows

Finding the pipelines your team already built, reading their inputs, estimating a run, and following it to the end.

Read the skill

Review

Judge a batch before it ships: scoring rules, verdicts, and reviewing a set against the brief.

layer-quality

Quality gates

Scoring generated files against a workspace's own rules, reading verdicts back, and choosing which candidate to deliver.

Read the skill

From a request to a usable asset

Ask in plain language. Your agent knows which model fits, what it costs, and what a good result has to satisfy before it hands one back.

Craft

The part a tool list cannot tell you

An OpenAPI description says which endpoints exist. It does not say that a pixel sprite needs a grid size in the prompt or it comes back anti-aliased, or that a tiling texture fails on baked-in lighting before it fails on the seam.

Every modality

Images, video, 3D, and audio

Concept art, sprites, textures, icons, marketing clips, game-ready meshes, sound effects and voice — and the timeline that cuts the shots into one file. The whole catalogue, not one corner of it.

Your art direction

A look that holds across a project

When to train a custom style on your own artwork rather than attach a reference, what to curate into it, and how to keep a cast on-model across a dozen images rather than a lucky one.

Spend

It asks before it spends

Every skill prices a run before executing it and stops for confirmation past a threshold. An agent left alone with a generation budget is the risk; this is the part of the pack that answers it.

Craft here, mechanics already published

Layer serves 6 more skills from the product itself. They cover authentication, connecting a client, the request shape, polling, training, and administration — REST first. The pack above does not restate any of it; it answers the questions they leave open. Install both.

The discovery index lists all 6 with a SHA-256 digest of the exact bytes each URL returns.

Agent Skills questions

What are Agent Skills?+
A skill is a Markdown document that teaches a coding agent a procedure — what to do, in what order, and what to check. Your agent loads the one that matches the work, so you do not have to explain Layer every time you ask for an asset. They follow the open Agent Skills convention.
How do I install them?+
Run npx skills add layerai/skills --skill "*" in your project. The --skill "*" matters: without it the command opens a picker with nothing preselected, so pressing enter installs nothing. In Claude Code you can instead add the pack as a plugin marketplace, which wires up the MCP server at the same time. Source at github.com/layerai/skills.
Do I need the MCP server as well?+
Yes — it is a prerequisite, not an alternative. The skills teach the procedure; the Layer MCP server is what carries it out. Add it once with OAuth and no API key, and the agent then knows what to call, in what order, and what to check.
How is this different from the skills Layer already publishes?+
Layer serves 6 skills from the product itself, at /.well-known/agent-skills/. Those teach mechanics — authentication, connecting a client, the request shape, polling, training, workspace administration. This pack is the craft layer above them: which model for the task, what the prompt must name, why a result is not landing. It does not restate the mechanics, and the two are designed to be installed together.
Which agents do they work with?+
The pack ships manifests for Claude Code, Cursor, Codex, VS Code, and the documents are plain Markdown, so any agent that reads SKILL.md can use them — and one that cannot can still be pointed at a file.
Do I need a Layer account?+
Yes — the skills describe how to work with Layer, and the work runs against your workspace. Signing up is self-serve and free; see self-serve access.
What does it cost to run?+
The pack is free and MIT-licensed. Generations your agent runs consume Creative Units from your workspace balance, exactly as they would in the app — no separate charge, no extra seat. The skills price a run before executing it and stop for confirmation past a threshold. See pricing.
Which models can my agent reach?+
All of them — 300+ across image, video, 3D, and audio, including day-zero integrations of new releases. The skills never hard-code a model id; they ask your workspace what it can run, so a new model is reachable the day it lands. Browse the catalogue.
Can I read them before installing?+
Every skill is a Markdown file in the public repository, so you can read exactly what your agent will — or vendor the ones you want and skip the installer.

Prefer to drive Layer yourself? The command-line interface covers the same ground from a terminal, and the REST API from your own code.

Put Layer in your agent's hands

One command adds the skills. Connect the MCP server and it can act on them.