Sudocode Nano Banana2

Generate images from text prompts or edit a local image with a prompt through the remote Sudocode API.

Initialize environment on first run

Before the first real use, check whether the required variables exist:

uv run python <skill_dir>/scripts/init_env.py --check-only

If the check reports missing variables, tell the user to first register at https://sudocode.us and apply for an API key, then run:

uv run python <skill_dir>/scripts/init_env.py

The initializer will prompt for values and persist them to ~/.openclaw/.env.

Use the bundled script

Run:

uv run python <skill_dir>/scripts/sudocode_nano_banana2.py --prompt "..." [--input_image path] [--output path]

Inputs:

• --prompt: required
• --input_image: optional local image path for image-to-image editing
• --output: optional output path, defaults to output.png

Environment:

• SUDOCODE_IMAGE_API_KEY: required API key; if missing, direct the user to register at https://sudocode.us and apply for an API key first
• SUDOCODE_BASE_URL: optional API base URL for the remote Sudocode service, defaults to https://sudocode.run

Behavior notes:

• This skill is an API client, not a local inference model
• It sends prompts and optional input images to the configured Sudocode endpoint
• It stores credentials in ~/.openclaw/.env only when the initializer is used
• It writes generated image files and per-run logs to local disk near the chosen output path

The initializer writes these values into ~/.openclaw/.env so future sessions can reuse them.

Remote API and model naming

This skill is named sudocode-nano-banana2 because it packages Sudocode's Nano Banana 2 image capability for OpenClaw.

The bundled script currently calls the upstream model identifier gemini-3.1-flash-image-preview exposed by the configured Sudocode-compatible endpoint. The skill name and upstream model identifier are intentionally different: one is the packaged skill name, the other is the current backend route used by the service.

If requests is missing, install it with:

uv pip install requests

Packaging and portability

This skill is structured as a standard skill package:

• SKILL.md: trigger metadata and operating instructions
• scripts/init_env.py: first-run environment bootstrap helper
• scripts/sudocode_nano_banana2.py: bundled runtime script

Do not include generated images, logs, or other local run artifacts in the distributable package.

Delivering generated images

When using this skill inside OpenClaw, send the generated file with the message tool so the user can actually receive the image.

Use:

message(action="send", filePath="/absolute/path/to/output.png")

Then reply NO_REPLY to avoid duplicate text.

Do not rely on reading the image file or pasting a local path into chat.

Error handling

If the script exits with code 2, relay this message to the user:

❌ 额度不足或请求过于频繁！
👉 请前往 https://sudocode.us 登录并充值后重试。

For other failures, show the script output directly.