# Skills

Install Atlas Cloud Agent Skills so Claude Code, Codex CLI, and Gemini CLI can generate images, video, audio, and 3D through the Atlas Cloud API.

Source: https://atlascloud.ai/docs/en/skills

## What are Skills?

Skills are reusable knowledge packages that teach an AI coding agent a specific domain: API references, parameter rules, model choices, and working code templates. Installing the Atlas Cloud skills gives your agent a precise understanding of the Atlas Cloud API, so it can generate images, video, audio, and 3D models — and write correct integration code — from plain-language instructions.

Atlas Cloud Skills work with **Claude Code**, **OpenAI Codex CLI**, **Gemini CLI**, and other agents that support the same skill format.

GitHub repository: [AtlasCloudAI/atlas-cloud-skills](https://github.com/AtlasCloudAI/atlas-cloud-skills)

## What's included

The repository ships **three skills**:

<Cards>
  <Card title="atlas-cloud" description="The core skill. Full API reference for image, video, audio, 3D, LLM chat, media upload, and model discovery." />
  <Card title="seedance-2-5-skill" description="Specialist skill for Seedance 2.5 — storyboarding, multimodal references, and native audio-video generation." />
  <Card title="universal-video-prompt-skill" description="Model-agnostic video prompting: turns a rough idea into a structured, shot-level prompt." />
</Cards>

### Inside the core skill

`atlas-cloud` is organized as a `SKILL.md` entry point plus seven reference documents that the agent loads on demand:

| Reference | Covers |
| --- | --- |
| `references/image-gen.md` | Image generation, including 3D models |
| `references/video-gen.md` | Video generation, lipsync, and talking avatars |
| `references/audio-gen.md` | Text-to-speech, music generation, and speech-to-text |
| `references/llm-chat.md` | OpenAI-compatible chat completions |
| `references/quick-generate.md` | One-step generation that picks the model for you |
| `references/upload.md` | Uploading local files for image-to-video and editing workflows |
| `references/models.md` | Model discovery, pricing, and capability lookup |

### Recipe library

Beyond the skills themselves, the repository includes **25 ready-made recipes** — end-to-end workflows an agent can follow:

| Category | Recipes | Examples |
| --- | --- | --- |
| `library/visual` | 7 | Product photography, poster design, style transfer |
| `library/motion` | 6 | Product render to ad, character to drama scene |
| `library/edit` | 6 | Background removal, object cleanup, upscaling |
| `library/social` | 6 | Long-form video to vertical shorts, thumbnail generation |

## Installation

<Tabs groupId="skill-install" items={['npx (recommended)', 'Shell script', 'Manual']}>
<Tab value="npx (recommended)">

Install all three skills:

```bash
npx skills add AtlasCloudAI/atlas-cloud-skills
```

Install a single skill:

```bash
npx skills add AtlasCloudAI/atlas-cloud-skills --skill atlas-cloud
```

</Tab>
<Tab value="Shell script">

Install all three skills:

```bash
curl -fsSL https://raw.githubusercontent.com/AtlasCloudAI/atlas-cloud-skills/main/install.sh | sh
```

Install a single skill by passing its name:

```bash
curl -fsSL https://raw.githubusercontent.com/AtlasCloudAI/atlas-cloud-skills/main/install.sh | sh -s atlas-cloud
curl -fsSL https://raw.githubusercontent.com/AtlasCloudAI/atlas-cloud-skills/main/install.sh | sh -s seedance-2-5-skill
curl -fsSL https://raw.githubusercontent.com/AtlasCloudAI/atlas-cloud-skills/main/install.sh | sh -s universal-video-prompt-skill
```

</Tab>
<Tab value="Manual">

Copy each skill directory into your agent's skill folder. For Claude Code that is `~/.claude/skills/`:

```bash
git clone https://github.com/AtlasCloudAI/atlas-cloud-skills.git
cp -r atlas-cloud-skills/atlas-cloud ~/.claude/skills/atlas-cloud
cp -r atlas-cloud-skills/skills/seedance-2-5-skill ~/.claude/skills/seedance-2-5-skill
cp -r atlas-cloud-skills/skills/universal-video-prompt-skill ~/.claude/skills/universal-video-prompt-skill
```

</Tab>
</Tabs>

## Setup

Set your API key as an environment variable:

```bash
export ATLASCLOUD_API_KEY="your-api-key"
```

To persist it across terminal sessions, add the line to your shell profile (`~/.bashrc`, `~/.zshrc`, and so on).

Create a key in the [Atlas Cloud dashboard](https://www.atlascloud.ai/console/api-keys). See [API Keys](/api-keys) for details.

<Callout type="warn">
Agents run shell commands on your behalf. Keep the key in your environment or a secrets manager — never commit it to a repository, and never paste it into a prompt.
</Callout>

## Usage examples

Once installed, describe what you want in plain language:

### Image and 3D

> "Generate a product photo of a ceramic mug on a linen backdrop with Seedream v5.0 Pro"

> "Turn this photo into a 3D model I can open in Blender"

### Video

> "Create a 5-second clip of ocean waves at golden hour using Seedance 2.5"

> "Take this product render and make a 15-second ad with a voiceover"

### Audio

> "Read this paragraph aloud in a warm female voice and save it as an MP3"

> "Write and generate a 90-second upbeat pop track about summer road trips"

> "Transcribe this interview recording and give me speaker-separated text"

### LLM and discovery

> "Ask DeepSeek V3.2 to review this file for security issues"

> "Which video models support reference images, and what do they cost?"

## Supported capabilities

| Capability | Endpoint | Example models |
| --- | --- | --- |
| Image generation | `POST /api/v1/model/generateImage` | Seedream v5.0 Pro, FLUX 2, Qwen Image 3.0 |
| 3D generation | `POST /api/v1/model/generateImage` | Hunyuan3D Pro, Seed3D 2.0, Tripo H3.1 |
| Video generation | `POST /api/v1/model/generateVideo` | Seedance 2.5, Kling, Wan 3.0, Veo |
| Speech synthesis | `POST /api/v1/model/generateAudio` | Seed Audio 1.0, ElevenLabs, MiniMax Speech |
| Music generation | `POST /api/v1/model/generateAudio` | Suno Chirp v5, MiniMax Music 3.0 |
| Speech-to-text | `POST /api/v1/model/generateAudio` | Seed ASR 2.0, xAI STT v1 |
| LLM chat | `POST /v1/chat/completions` | DeepSeek, Qwen, Kimi, GLM, MiniMax |
| Media upload | `POST /api/v1/model/uploadMedia` | Local file to hosted URL |
| Model discovery | Catalog lookup | Search by name, type, or provider |

<Callout>
3D generation runs through the image endpoint, and speech, music, and transcription all share the audio endpoint. The direction is determined by the model you pick, not by the URL. See [Audio models](/models/audio) and [Predictions](/predictions) for the full request and polling flow.
</Callout>

## Skills or MCP Server?

Both bring Atlas Cloud into your development environment, but they solve different problems:

| | Skills | MCP Server |
| --- | --- | --- |
| Works with | Claude Code, Codex CLI, Gemini CLI, and other skill-aware agents | Claude Code, Cursor, Claude Desktop, VS Code, Windsurf, Cline, and more |
| What it provides | Domain knowledge, parameter rules, and code templates | Executable tools the agent calls directly |
| Setup | One command | JSON or TOML config per client |
| Best for | Writing integration code, multi-step creative workflows | Running generations without leaving the editor |

**Use both.** The MCP Server executes the calls; the skills teach your agent which model to pick and which parameters actually matter.

See the [MCP Server](/mcp-server) guide for setup.

## Using skills with DeepSeek Harness

If you work in DeepSeek Harness, the [dsh-media-gen](https://github.com/AtlasCloudAI/dsh-media-gen) bundle packages these skills as a profile plugin:

```sh
dsh plugin --profile web add 'github:AtlasCloudAI/dsh-media-gen#v0.2.0'
export ATLASCLOUD_API_KEY="your-api-key"
```

The bundle tracks the upstream skills and can optionally enable the Atlas Cloud MCP server alongside them.

## Related

<Cards>
  <Card title="MCP Server" href="/mcp-server" description="Run Atlas Cloud tools directly from your IDE." />
  <Card title="CLI" href="/cli" description="Generate and manage jobs from the terminal." />
  <Card title="API Keys" href="/api-keys" description="Create and scope the key your agent will use." />
</Cards>