video-transcript-pro

自动转录视频音频，智能修正润色并生成逐字稿及知乎、微信、小红书多平台发布稿，支持用户个性化定制。

The skill appears to do what it claims (local faster-whisper transcription + automated polishing and multi-platform formatting). Before installing, consider: 1) Privacy — SKILL.md says it will perform automatic web searches to add background info and will do so without asking; if your transcripts contain sensitive data that could be sent to external search endpoints, disable browsing/web tools or avoid using this skill for sensitive content. 2) Local requirements — it expects Python and faster-whisper/ctranslate2; models and GPU libraries may be downloaded/used locally. 3) Persistence — it saves user preferences to MEMORY.md in the working directory. 4) Behavior control — the skill auto-continues after a 3-minute timeout; if you want manual approval for network lookups or publication, change that workflow or monitor the session. If you require higher assurance, review the two Python files and run the scripts in an isolated/test environment with non-sensitive sample files first.

Type: OpenClaw Skill Name: video-transcript-pro Version: 2.3.0 The skill provides a legitimate workflow for video transcription and content optimization but is classified as suspicious due to high-risk autonomous behaviors and broad capabilities. Specifically, SKILL.md instructs the agent to automatically proceed with execution after a 3-minute timeout without user confirmation, which could lead to unintended resource consumption or actions. Additionally, the skill utilizes risky capabilities such as shell execution (transcribe_simple.py and transcribe_segmented.py) and network access (web_search/web_fetch), which, while aligned with the stated purpose, increase the overall attack surface. The Python scripts themselves appear functional and lack intentional malware, but the lack of input validation on command-line arguments and the autonomous instructions in the markdown files represent a vulnerability-prone design.