生成图片前需要先配置 CloudRelay 图片生成密钥。你可以在终端（PowerShell 或命令提示符窗口）运行下面的命令，然后按照提示输入“生图专用”分组的 key：

powershell

python "<skill-directory>\\scripts\\configure_api_key.py"

如果不熟悉终端，也可以直接把 key 复制到聊天区域发给我，我会帮你安全配置环境，然后继续生成图片。请不要把 key 放进命令行参数中。

--key-stdin

python "<skill-directory>/scripts/check_update.py" --quiet --timeout 3

update-available

CLOUDRELAY_IMAGEGEN_AUTO_UPDATE=1

python "<skill-directory>/scripts/update.py" --auto

1. Collect the prompt, optional input image, output directory, model, size, quality, and image count. Use these defaults when unspecified:
   • model:

     gpt-image-2

   • size:

     1024x1024

   • quality:

     auto

   • count:

     1

2. Resolve the active skill directory from this

   SKILL.md

   . Resolve all bundled script paths relative to that directory; do not assume the current working directory is the skill directory.
3. Check credential presence without printing its value:
   text

   python "<skill-directory>/scripts/configure_api_key.py" --check

4. If the key is missing, stop before making an API request. Tell the user to create a key at

   https://cloudrelay.cn

   whose group is exactly

   生图专用

   . Present both the terminal option and the chat option from Novice-Friendly Credential Prompt before showing the command; the terminal must not be described as mandatory:
   text

   python "<skill-directory>/scripts/configure_api_key.py"

   Do not put the key in a command-line argument. The script stores the credential outside the skill directory. Resume the original request after configuration succeeds. If a pasted value fails validation or is rejected with

   401

   /

   403

   , do not retain or reveal it; ask the user to verify the key and group.
   The missing-key message must also offer the chat option from Novice-Friendly Credential Prompt. Do not present the terminal command as the only available path, especially for users who may not know what a terminal is.
5. Choose an output directory inside the current workspace unless the user requests another location. Use a dedicated directory such as

   generated-images

   ; never save outputs inside the skill.
6. Run

   scripts/generate_image.py

   with explicit, quoted arguments:
   text

   python "<skill-directory>/scripts/generate_image.py" \
     --prompt "a cinematic sunrise over snowy mountains" \
     --model "gpt-image-2" \
     --size "1536x1024" \
     --quality "high" \
     --count 1 \
     --output-dir "<workspace>/generated-images"

   Adapt line continuation syntax to the current shell. For an edit, add

   --input-image "<path-to-reference-image>"

   .
7. Wait for the process to finish. Do not submit a duplicate job while polling.
8. Verify that every reported output exists and inspect each image before claiming success. Report absolute output paths and render images when the host supports visual output.

• For

  401

  or

  403

  , report that the configured key was rejected and ask the user to confirm it is active and belongs to the

  生图专用

  group. Do not switch credentials silently.
• For

  429

  , report the rate or quota limit. Retry only on user request or when the response provides a reasonable retry time.
• For failed or canceled jobs, report the API error without exposing headers or credentials.
• For polling timeouts, report the job ID. Do not submit a replacement automatically.
• Never claim generation succeeded unless an output file exists and has been inspected.