# ViewMax Agent Skills

Install ViewMax Agent Skills for credit-aware video, image, music, and audio generation through the REST API.

Agent Skills teach an agent to call the ViewMax REST API with `curl`. Prefer [MCP](https://viewmax.studio/docs/mcp) when your client supports it; Skills suit agents that can run shell commands but have no MCP connection.

## Install

Download the video Skill and its concise API reference into a Claude skills directory:

```bash
curl -fsSL https://viewmax.studio/skills/viewmax-video/SKILL.md \
  --create-dirs -o ~/.claude/skills/viewmax-video/SKILL.md
curl -fsSL https://viewmax.studio/skills/viewmax-video/reference.md \
  --create-dirs -o ~/.claude/skills/viewmax-video/reference.md
```

To share a Skill with a project, download it to `.claude/skills/<name>/` in the repository instead. Other agents that load Agent Skills can use the same two files from their own skills directory.

The Skill is named `viewmax-video`. Its trigger description covers requests to generate video, use Seedance, Veo, Kling, or MiniMax, or turn an image into video.

## Workflow

The installed Skill instructs an agent to:

1. Check that `VIEWMAX_API_KEY` is available.
2. Fetch `GET https://viewmax.studio/api/v1/models` and choose a current model, its allowed values, and `max_prompt_length`.
3. Tell the user the model and credit cost, and wait for confirmation before creating a task.
4. Create the task with `POST /api/v1/videos` and an `Idempotency-Key` header, reusing the key on a retry.
5. Poll `GET /api/v1/tasks/{taskId}` every 5–10 seconds until `success`, `failed`, or `canceled`.
6. Return the first `data.taskUrls` entry after success, or report `errorMessage` from `data.taskInfo` after a failure.

`reference.md` contains the endpoint table, accepted request fields, task lifecycle, and the error types with the action for each.

## Image, music, and audio skills

The same pattern is available for the other modalities. Each Skill triggers on its media type and wraps the matching v1 endpoint.

```bash
# Image (text-to-image and image-to-image)
curl -fsSL https://viewmax.studio/skills/viewmax-image/SKILL.md \
  --create-dirs -o ~/.claude/skills/viewmax-image/SKILL.md
curl -fsSL https://viewmax.studio/skills/viewmax-image/reference.md \
  --create-dirs -o ~/.claude/skills/viewmax-image/reference.md

# Music
curl -fsSL https://viewmax.studio/skills/viewmax-music/SKILL.md \
  --create-dirs -o ~/.claude/skills/viewmax-music/SKILL.md
curl -fsSL https://viewmax.studio/skills/viewmax-music/reference.md \
  --create-dirs -o ~/.claude/skills/viewmax-music/reference.md

# Audio (text-to-speech and sound effects)
curl -fsSL https://viewmax.studio/skills/viewmax-audio/SKILL.md \
  --create-dirs -o ~/.claude/skills/viewmax-audio/SKILL.md
curl -fsSL https://viewmax.studio/skills/viewmax-audio/reference.md \
  --create-dirs -o ~/.claude/skills/viewmax-audio/reference.md
```

- `viewmax-image` — text-to-image and image-to-image via `POST /api/v1/images`.
- `viewmax-music` — original music tracks via `POST /api/v1/music` (60 credits up to 60 seconds, then 1 credit per second).
- `viewmax-audio` — text-to-speech via `POST /api/v1/speech` (`ceil(characters / 20)` credits, at least 1) and sound effects via `POST /api/v1/sound-effects` (5 credits).

## Credentials and safety

The Skills send `Authorization: Bearer $VIEWMAX_API_KEY`. Keep the key in the agent's secure environment, not in a prompt or client-side source. A Skill must not create a billable task before the user confirms the credit cost.