菜单
Use when the user asks AI Canvas to generate, create, fill, or design an image into the current canvas holder, including phrases like 填充图片框, 生成广告图, 做一张海报, or Fill AI Image Holder.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Use when the user wants AI Canvas to process Skill jobs submitted from the canvas Skill panel, including 继续处理 Skill, 处理画布 Skill, 小红书封面 Skill, or canvas skill requests.
Use when the user wants an end-to-end natural-language AI infinite canvas workflow: opening AI Canvas, generating images into holders, reading canvas annotations, handling phrases like 要求后续变更 / 按标注修改, and creating edited image versions.
Use when the user wants AI Canvas to automatically process edit jobs submitted from the canvas button, including 开启自动修图模式, 继续自动修图, 自动监听画布修改, 点按钮后自动修图, or continuous AI Canvas annotation edit mode.
Use when the user has annotated an AI Canvas image and asks for follow-up changes, including 要求后续变更, 后续修改, 按标注修改, 根据箭头修改, 照画布意见改图, or edit the selected image from canvas annotations.
| name | fill-ai-image-holder |
| description | Use when the user asks AI Canvas to generate, create, fill, or design an image into the current canvas holder, including phrases like 填充图片框, 生成广告图, 做一张海报, or Fill AI Image Holder. |
Use this skill for the first image-generation loop in AI Canvas. The user should be able to say a normal request, not tool names.
This workflow must use the AI Canvas MCP tools. If prepare_image_generation, insert_image_into_holder, or save_snapshot are not callable in the current Codex thread, stop and tell the user:
AI Canvas 插件已经识别到，但工具没有加载出来。如果你刚安装或刚更新过插件，请完全退出并重新打开 Codex，再发送同一句需求；插件不需要重新安装。
Do not inspect plugin files, run curl, check ports, start local services by hand, or recreate HTTP calls during a normal image-generation request. Those actions are only for explicit plugin development/debugging.
prepare_image_generation with the user's original request.16:9.5:7.1:1.5:7.open, or Browser automation to connect the canvas.suggestedPrompt. If the user asked for text, typography, title design, poster lettering, or brand words, let image 2.0 design those directly in the image.insert_image_into_holder with the holder id and local image path.save_snapshot.auto-ai-canvas-edit-mode workflow by polling watch_edit_requests until the user says to stop, the Codex app/thread is closed, the task is interrupted, or a blocking clarification is required.Say what is happening like a product assistant:
Do not expose MCP JSON, shape IDs, or file paths unless the user asks for debugging details.
Do not launch an external browser window, hidden Chrome, headless Chrome, shell open, or Browser automation. Show the returned local URL as a clickable link and let the user open it inside Codex's side panel. Browser automation is for testing only.
If the canvas page is not open, keep the generation flow simple: generate the image, save it, insert it through the AI Canvas service, and let the service queue the visual update until the user opens the canvas.