Agent Skills
› KunAgent/Kun
› image-generation
image-generation
GitHub使用Kun工具生成或精修单张图像，明确构图并引用授权素材。流程涵盖需求解析、比例选择、参考图关联及结果校验，确保符合身份与风格要求。
触发场景
需要生成指定风格的图像
需要对已有图像进行精修或编辑
安装
npx skills add KunAgent/Kun --skill image-generation -g -y
SKILL.md
Frontmatter
{
"name": "image-generation",
"description": "Generate or refine one image with Kun's image tool using explicit composition and authorized references."
}
Image Generation
Copyright (c) 2026 KunAgent. Licensed under the MIT License.
Purpose
Generate or refine one image with Kun's image tool using explicit composition and authorized references.
Tool routing
| Tool or skill | Use |
|---|---|
generate_image |
Generate exactly one image per call. |
read |
Inspect authorized workspace references when needed. |
Workflow
- Translate the request into subject, composition, viewpoint, style, lighting, palette, text requirements, and exclusions.
- Choose an aspect ratio that matches the destination.
- Attach only authorized current-thread or workspace reference images.
- Generate one image.
- Inspect the result and refine only when necessary or requested.
Completion gates
- Set image_size only when the user explicitly requests 1K or 2K.
- Check faces, hands, text, logos, geometry, and cropping where relevant.
- Preserve the requested identity and composition when editing from a reference.
Boundaries
- Do not claim ownership of third-party styles, logos, or characters.
- Do not copy temporary attachments into the workspace.
Delivery
Lead with the outcome, name the evidence used for verification, and disclose any real limitation that remains.
版本历史
- a984912 当前 2026-09-09 01:50
