Nano Banana Cut 图片生成切割，用于短视频创作，解决角色一致性问题和故事叙事

AI图片生成与智能切割工具，基于AceData Nano Banana模型，支持多分辨率多尺寸生成，自动切割为2/4/6/9宫格，自带瀑布流作品管理、批量下载功能。使用场景：(1) 输入prompt生成AI图片并自动切割成九宫格等多宫格 (2) 上传图片进行智能多宫格切割 (3) 管理生成的图片作品，支持打包下载

What to consider before installing: - The skill requires an API_KEY (mandatory) and optionally a PLATFORM_TOKEN (for uploads). The registry metadata did not declare these; expect to provide them and the skill will save them into a local .env file in the skill folder. Do not use high-privilege or long-lived credentials without confirming the remote service and the token scope. - The package runs a local Flask web server on port 697, writes files (images, thumbnails, a SQLite DB) and may upload files to platform.acedata.cloud. If you host sensitive images, review upload behavior and the remote endpoint policy. - Inspect the code yourself (server.py, task.py, upload.py, cut.py) before running. Pay attention to any hard-coded paths (set.json contains a user-specific save_path) and to logging that may echo request/response bodies to the console. - The pre-scan flagged a potential base64 prompt-injection pattern in the SKILL.md; search the docs/files for embedded or obfuscated data blocks before trusting the skill. - If you want to try it: run it in an isolated environment (container or VM), provide a scoped API key, and monitor outgoing network traffic. If you can't audit the code or do not trust the external AceData endpoints, do not provide real credentials or avoid installing it. - If you need higher confidence, ask the publisher for: (1) the canonical upstream source/homepage for the project, (2) confirmation of what exact token scopes are needed, and (3) a version of set.json without user-specific paths and without trailing/invalid JSON syntax so the code's config-loading behavior can be validated.

Type: OpenClaw Skill Name: nano-banana-cut Version: 1.0.0 The skill bundle provides a functional AI image generation and processing tool, but contains significant security vulnerabilities. Specifically, the `serve_file` route in `server.py` is vulnerable to path traversal, potentially allowing arbitrary file reads from the host system by joining user-provided paths with the root directory. Additionally, the `open_folder` endpoint in `server.py` uses `os.startfile` on paths retrieved from the database without sufficient validation, and the configuration in `set.json` includes hardcoded local file paths specific to the developer's environment (e.g., `C:/Users/86137/Desktop/banana`). While these appear to be unintentional security flaws rather than deliberate malware, they pose a risk to the host environment.