video-edit-strategy

分析素材与用户意图，输出结构化 JSON 剪辑策略（分镜、时间线、转场、音频、文字）。当用户要求制作短视频、混剪、或提供了素材但未给出具体剪辑指令时调用。策略输出供 ffmpeg-cli / ffmpeg-video-editor 等下游 skill 执行。

This skill appears to do what it says: create a structured JSON editing plan for provided media. Before installing or invoking it, consider: 1) The SKILL.md expects ffprobe/ffmpeg tooling but the metadata did not list required binaries — ensure your environment actually has ffprobe/ffmpeg or steps that call them will fail. 2) The skill will probe any file paths you provide (ffprobe reads local files) and will include those original paths in the generated plan; downstream execution may then read those files. Only provide paths to files you are comfortable allowing the agent and downstream skills to reference. 3) Intermediate artifacts are placed under /tmp/ve_strategy — check and clean that directory if needed. 4) The skill does not request credentials or external endpoints, but downstream executor skills (ffmpeg-cli, ffmpeg-video-editor) will perform actual file operations; only use trusted executors. 5) If you need stricter sandboxing, run this skill in an environment with limited filesystem access or only provide copies of needed files. If you want higher assurance, ask the publisher/source (not provided here) to declare required binaries and to confirm expected downstream skill mappings.

Type: OpenClaw Skill Name: video-edit-strategy Version: 1.0.0 The skill is designed to generate structured video editing strategies but contains a potential shell injection vulnerability in its core workflow. In SKILL.md, the agent is instructed to execute a shell command (`ffprobe`) using a template that wraps the user-provided 'INPUT_FILE' in quotes without explicit instructions for input sanitization, which could be exploited if a filename contains command-injection sequences. While the behavior is aligned with the stated purpose of video orchestration, the lack of security boundaries when interacting with the shell and the generation of complex execution plans for downstream skills (like ffmpeg-cli) qualify it as suspicious under the vulnerability criteria.