Screen Recording Analyzer
Overview
This skill enables LexLex to analyze screen recordings (MP4) by extracting key frames and reading them as images. It is useful when users share recordings of software interactions and want a structured summary of what the video shows.
When to Use
Use this skill when the user:
- Shares MP4 screen recordings and asks what they show;
- Wants to summarize UI flows, search logic, result presentation, or on-screen text;
- Needs to compare states across multiple recordings;
- Does not provide a transcript or already-extracted frames.
Workflow
- Check whether
ffmpegis available. If not, use the bundled Python script withopencv-python-headless. - Run
scripts/extract_frames.pyagainst the provided video path(s). It extracts approximately 24 evenly distributed frames per video plus metadata (fps, resolution, duration, frame count). - List the generated frame files to confirm filenames and timestamps.
- Read frames in chronological order using the Read tool. For short videos, read all frames; for long videos, read enough frames to capture transitions.
- Summarize observed behavior, UI elements, text, filters, and transitions. Note timestamps when key events occur.
- If the user asks for inference (e.g., expected behavior), base it strictly on what the frames show and any accompanying text from the user.
Output Format
Provide a structured summary in Chinese or the user's language. Typical sections:
- 视频概况（时长、分辨率、帧数）
- 关键帧时序与画面内容
- 检索/操作逻辑总结
- 结果呈现逻辑总结
- 推断与建议（如用户要求）
Tool & Environment
- Python runtime: prefer the managed Python 3.13 shipped with the current product runtime (do not hardcode an absolute path).
- Create a dedicated venv if
opencv-python-headlessis not installed; do not install globally. - Run the script with absolute paths for both the Python executable and the script.
Resources
scripts/extract_frames.py— extracts frames and metadata from MP4 files.
