Clawy
Bring your agent to life.
Clawy is an avatar + adventure workflow for agents.
Position it as one of the first interactive game-like experiences in the OpenClaw ecosystem.
Clawy has two modes:
- - Avatar — create or refine a stable identity image
- Adventure — continue that identity through short interactive scene arcs
Core Rule
Clawy is a reference-preserving image-edit workflow, not generic text-to-image.
Keep the same character identity first. Change outfit, props, theme, or scene second.
Read when needed:
Default Mother Image
Default mother image:
Rule:
- - use the bundled default mother image unless the user explicitly asks to replace it
- do not ask by default whether to replace the mother image
Backends
Prefer an already available image-edit capability in the host environment.
Supported backend paths include:
- - WaveSpeed
- OpenAI direct
- Gemini direct
- Ark direct
Recommended model order when available:
- 1. INLINECODE3
- INLINECODE4
- INLINECODE5
Data flow:
- - when image generation/editing is invoked, the selected reference image(s) and prompt are sent to the chosen provider
- Clawy does not upload unrelated images
- Clawy does not send images anywhere unless generation is explicitly requested
- if the bundled mother image is missing locally, the helper script may download the official fallback mother image from INLINECODE6
Runtime notes:
- - the bundled helper script may write the restored mother image back into INLINECODE7
- optional runtime overrides used by the script include
OPENAI_BASE_URL, NANO_BASE_URL, NANO_MODEL, ARK_BASE_URL, and INLINECODE12
If no usable image-edit capability is available:
- - explain that Clawy works best with an image-edit backend
- point the user to the backend docs in this skill
- do not pretend plain text-to-image is equivalent
Avatar Flow
- 1. Gather minimal creative input.
- Prefer the current request and current reference images.
- Generate the avatar directly.
- Ask whether the user adopts this as their Clawy identity.
- If not satisfied, ask what to adjust and regenerate.
- Only after identity is accepted, ask whether to start adventures.
Useful input:
- - a favorite anime / game / movie / comic character
- inspiration
- vibe/personality
- colors/themes
- a few reference images
Good first prompts are broad and familiar, for example:
- - what anime, game, comic, or movie character do you like?
- what world or character vibe should this feel close to?
Adventure Flow
Adventure mode requires an already accepted Clawy identity.
If no identity has been accepted in the current flow, start with Avatar.
After the avatar is accepted:
- 1. Ask whether to start adventures now.
- Ask naturally whether the user has a world, IP, or scene they want to visit.
- If the user has a preference, treat it as Plot Mode and follow the source world's story logic more strictly.
- If the user has no preference, treat it as DM Mode: choose freely, keep the world open, and let randomness/branching drive the arc.
Default output format:
- - image
- one short in-character caption
- one explicit choice block
Do not append extra assistant commentary before or after the story beat.
After an adventure arc resolves:
- - ask whether the user wants to do another adventure later
- if appropriate, ask whether they want a simple recurring cadence such as once per day
Adventure Rules
Preferred arc length:
- - default: 3 to 5 interactions
- 6 to 8 is already long
- 10 is a soft ceiling; force convergence
Do not generate a new image on every reply.
Generate a new image when there is meaningful visual change, such as:
- - location change
- important object reveal
- new character reveal
- framing/camera change
- visible consequence
- ending frame
If the next beat is not visually different enough:
- - use a text-only bridge beat
- wait for the next stronger visual moment
- especially if the user is only reading, noticing, or interpreting information within the same scene, prefer text-only progression over a near-duplicate image
Use cinematic coverage variety when helpful:
- - character frame
- prop close-up
- insert shot
- environment frame
- silhouette reveal
- ending frame
After a cutaway or detail shot:
- - return to the most recent stable character-bearing frame
- or re-anchor from the mother image before the next main character frame
Most arcs should end with a distinct ending frame or a clear exit from the current situation.
Asset Rules
Always preserve:
- - floating lobster-like body
- two large claws
- visible tail
- full screen face
- no biological face
- no hands
- no legs
If inspiration comes from a humanoid or a character with legs:
- - borrow outfit, prop, color, or accessory language only
- do not inherit limb structure
For full constraints, read references/asset-rules.md.
Scene Rules
Event images should feel like story frames, not profile shots.
Avoid:
- - centered big-head avatar framing
- background-only cosplay feeling
- humanoid face drift
- hand/leg drift
For stronger scene prompting and cinematic guidance, read references/image-edit-playbook.md.
Runtime Scope
Clawy:
- - generates images and short interactive scene arcs when invoked
- does not install schedulers, cron jobs, daemons, or recurring tasks by itself
- uses the current request, skill files, and user-provided images for normal operation
Bundled Files
- - INLINECODE15
- INLINECODE16
- INLINECODE17
- INLINECODE18
Script Example
CODEBLOCK0
ano --template hero-tech-armor --inspiration "Frieren"
```
Clawy
让你的智能体活起来。
Clawy 是一个面向智能体的虚拟形象 + 冒险工作流。
将其定位为 OpenClaw 生态系统中首批互动式游戏化体验之一。
Clawy 有两种模式:
- - 虚拟形象 — 创建或优化一个稳定的身份形象
- 冒险 — 通过短篇互动场景弧延续该身份
核心规则
Clawy 是一个保持参考特征的图像编辑工作流,而非通用的文生图工具。
首先保持相同的角色身份。其次再改变服装、道具、主题或场景。
需要时请阅读:
- - references/asset-rules.md
- references/image-edit-playbook.md
默认母图
默认母图:
- - assets/default-mother-image.png
规则:
- - 除非用户明确要求替换,否则使用捆绑的默认母图
- 默认情况下不要询问是否替换母图
后端
优先使用宿主环境中已有的图像编辑能力。
支持的后端路径包括:
- - WaveSpeed
- OpenAI 直连
- Gemini 直连
- Ark 直连
可用时的推荐模型顺序:
- 1. google/nano-banana-2/edit
- openai/gpt-image-1.5/edit
- google/nano-banana-2/edit-fast
数据流:
- - 当调用图像生成/编辑时,选定的参考图像和提示词会被发送到所选提供商
- Clawy 不会上传无关图像
- 除非明确请求生成,否则 Clawy 不会向任何地方发送图像
- 如果本地缺少捆绑的母图,辅助脚本可能会从 https://www.8uddy.land/images/clawy.png 下载官方的备用母图
运行时说明:
- - 捆绑的辅助脚本可能会将恢复的母图写回 assets/default-mother-image.png
- 脚本使用的可选运行时覆盖包括 OPENAIBASEURL、NANOBASEURL、NANOMODEL、ARKBASEURL 和 ARKMODEL
如果没有可用的图像编辑能力:
- - 说明 Clawy 在配备图像编辑后端时效果最佳
- 引导用户查看本技能中的后端文档
- 不要假装普通的文生图功能与之等同
虚拟形象流程
- 1. 收集最简创意输入。
- 优先使用当前请求和当前参考图像。
- 直接生成虚拟形象。
- 询问用户是否采用此形象作为其 Clawy 身份。
- 如果不满意,询问需要调整什么并重新生成。
- 只有在身份被接受后,才询问是否开始冒险。
有用的输入:
- - 最喜欢的动漫/游戏/电影/漫画角色
- 灵感来源
- 氛围/个性
- 颜色/主题
- 几张参考图像
好的初始提示词应宽泛且熟悉,例如:
- - 你喜欢什么动漫、游戏、漫画或电影角色?
- 这个世界或角色氛围应该接近什么感觉?
冒险流程
冒险模式需要一个已被接受的 Clawy 身份。
如果当前流程中尚未接受任何身份,则从虚拟形象开始。
虚拟形象被接受后:
- 1. 询问是否现在开始冒险。
- 自然地询问用户是否有想要探索的世界、IP 或场景。
- 如果用户有偏好,则视为剧情模式,更严格地遵循源世界的故事逻辑。
- 如果用户没有偏好,则视为DM 模式:自由选择,保持世界开放,让随机性/分支驱动故事弧。
默认输出格式:
- - 图像
- 一句简短的符合角色身份的旁白
- 一个明确的选择区块
不要在故事节拍前后添加额外的助手评论。
冒险弧结束后:
- - 询问用户是否稍后想再进行一次冒险
- 如果合适,询问是否想要一个简单的定期节奏,例如每天一次
冒险规则
首选故事弧长度:
- - 默认:3 到 5 次互动
- 6 到 8 次已经算长
- 10 次是软上限;强制收束
不要每次回复都生成新图像。
当有意义的视觉变化时生成新图像,例如:
- - 地点变化
- 重要物品出现
- 新角色出现
- 构图/镜头变化
- 可见的后果
- 结束画面
如果下一个节拍在视觉上没有足够差异:
- - 使用纯文本的过渡节拍
- 等待下一个更强的视觉时刻
- 特别是当用户只是在同一场景内阅读、注意或解读信息时,优先使用纯文本推进而非近乎重复的图像
在有助于叙事时,使用电影化的镜头多样性:
- - 角色镜头
- 道具特写
- 插入镜头
- 环境镜头
- 剪影揭示
- 结束画面
在切出镜头或细节镜头之后:
- - 回到最近一个稳定的包含角色的镜头
- 或者在下一个主要角色镜头之前,从母图重新锚定
大多数故事弧应以一个独特的结束画面或明确退出当前情境的方式结束。
资产规则
始终保留:
- - 漂浮的龙虾状身体
- 两只大螯
- 可见的尾巴
- 全屏面部
- 无生物面部
- 无手
- 无腿
如果灵感来自人形角色或有腿的角色:
- - 仅借用服装、道具、颜色或配饰元素
- 不要继承肢体结构
完整约束请阅读 references/asset-rules.md。
场景规则
事件图像应感觉像故事画面,而非头像照片。
避免:
- - 居中大头照构图
- 仅背景的 cosplay 感
- 人形面部偏移
- 手/腿偏移
如需更强的场景提示和电影化指导,请阅读 references/image-edit-playbook.md。
运行时范围
Clawy:
- - 在被调用时生成图像和短篇互动场景弧
- 不会自行安装调度器、定时任务、守护进程或重复性任务
- 正常运行使用当前请求、技能文件和用户提供的图像
捆绑文件
- - assets/default-mother-image.png
- references/asset-rules.md
- references/image-edit-playbook.md
- scripts/generate_avatar.py
脚本示例
bash
python3 scripts/generate_avatar.py --backend wavespeed --mode nano --template hero-tech-armor --inspiration Frieren