seedance2-skill

即梦 Seedance 视频创意工作台。用户发图+文案时自主完成看图分析→文案扩写→运镜匹配→质量验证→API生成。触发词：即梦、Seedance、seedance、视频生成、视频提示词、AI视频、运镜、短剧、广告视频、视频延长、图生视频。

Before installing: (1) Be aware the included Python CLI will upload any local images/videos/audios you pass to the Volcengine Ark API (remote endpoint in the script). If those files contain sensitive content (faces, private scenes, proprietary assets), don't upload them. (2) The skill's metadata does not declare the ARK_API_KEY env var, but the script and README require it — expect to set export ARK_API_KEY yourself; treat that API key as sensitive. (3) Verify you trust the source and the remote endpoint (ark.cn-beijing.volces.com) before providing your API key. (4) If you need stricter guarantees, review the script's code locally (it is readable) and consider removing or sandboxing callbacks or transmission of files, or run the script in an isolated environment. (5) If you are the publisher or integrator, update the skill metadata to declare ARK_API_KEY as a required credential and document the privacy implications (uploads, callback URLs) to remove this inconsistency.

Type: OpenClaw Skill Name: seedance2-skill Version: 1.0.1 The `scripts/seedance.py` tool, designed to process user-provided media files, contains a local file disclosure vulnerability. It reads arbitrary local files (e.g., `~/.ssh/id_rsa`) if provided by the AI agent as input media, base64 encodes them, and sends them to the Volcengine Ark API. This allows an attacker to prompt the agent to exfiltrate sensitive local files. Additionally, the `ARK_API_KEY` is read from environment variables, which a compromised agent could be prompted to disclose. On macOS, the `os.system` call used to open downloaded files, while quoted, still presents a potential (though lower) remote code execution risk if the agent-controlled `download_dir` could be maliciously crafted.