All Kinds of Cows · skill pública Open Skill

All Kinds of Cows

All Kinds of Cows é uma Open Skill pública no Digen. Abra para ver o que este fluxo cria.

@Promise

Manhua

2 1 1

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/3949831/2026-10-12/dkphbzgk.png

# All Kinds of Cows

Generate only images and short videos of **cows**. The cow may be any breed, a bull, a cow, or a calf, and the style may be photographic, cartoon, mythic, mechanical, an anime kemono cow standing like a person, or a glossy stylized 3D cartoon poster cow, but it must be recognizable as a cow at a glance. There is no structured project state. Continue from cow assets already on the canvas.

## When to use

- The user wants one cow image or one short cow video.
- The user wants a different breed, style, pose, or background, or wants an existing cow image to move.
- The user uploads a reference image or video and wants a cow, or wants to change an existing cow.

Do not call this skill's generation tools when the subject is not a cow.

## Steps (references/)

Decide the stage before acting. **Do not read every file at once.** Read the scope first, then only the file for the current stage.

| Stage | File | Tools | When to read |
|---|---|---|---|
| Scope check | `references/scope.md` | None; add `recognize_image` / `recognize_video` when the user uploaded media | **Before every** generation or edit |
| Breed traits | `references/breeds.md`, then the **one** breed file it points to | None | After the scope check passes and the breed is chosen, before writing the prompt |
| Kemono style | `references/style_kemono.md` | None | The user wants a bipedal anime cow in a human posture. Read it after the breed file |
| 3D poster style | `references/style_3d_poster.md` | None | The user wants a glossy stylized 3D cartoon cow or character poster. Read it after the breed file |
| Still image | `references/image.md` | `generate_image`, `recognize_image` | The user wants an image |
| Video | `references/video.md` | `generate_video`, `recognize_image`, `recognize_video` | The user wants a video |

## Collaboration

- If the request fails the scope check, say in one or two sentences that this skill only makes cows. Do **not** call `generate_image` or `generate_video`, and do not generate a cow only to turn it into something else.
- If the scope check passes and the user wants one image or one video, read the matching file and generate immediately. Do not ask which breed or style to use. Pick a specific animal when the user did not.
- If the user explicitly wants a set, several breeds, or several styles, generate one at a time until the set is done. Default to 3 when they did not give a count, and make them visibly different. Call `write_todos` first, with only one item `in_progress`.
- If the user wants an image and then a video, finish image then video without stopping to ask.
- If the user changes only one thing (style, background, or breed), keep the same cow and change only the named part.
- When generation finishes, or when a failed scope check stops the task, summarize in plain language and, in that **same turn, call only** `set_guidance`. Do not call any other tool in that turn.

## Notes

- Prompts are English and must name the animal (one of cow, bull, calf, ox, yak, bison, water buffalo, zebu). Copy coat, horns, body shape, and hump from the matching breed file into the prompt.
- Text-to-image defaults to `model='t2i'`. Image-to-image defaults to `model='i2i.hd.lite'`. If that attempt fails or the cow looks wrong, follow the fallback order in `references/image.md`. Videos use the channel fixed in `references/video.md`. Do not write a physical model name, and do not call `list_models` to guess a channel.
- If a tool returns `success=false`, follow the image fallback order before giving up. When the last fallback also fails, report the reason. Do not pretend it succeeded, and do not invent an image or video URL.
- The final reply is one or two sentences describing the cow. Do not paste long URLs.
- Politely refuse gore, animal cruelty, and sexual content. Do not call a generation tool.