video-pipeline

Video Pipeline is a comprehensive automation agent skill designed to streamline the lifecycle of video content processing. It acts as an orchestrator for a suite of specialized skills, moving from raw video files (such as live streams, podcast recordings, or conference footage) to a professional-grade multi-format output. This solution is specifically designed for creators, marketing teams, and knowledge workers who need to repurpose long-form audiovisual media into diverse content assets without manual intervention.

• Orchestrates a four-stage pipeline: transcript generation, intelligent insight extraction, automated clip generation, and article composition.

• Leverages WhisperX for high-accuracy speech-to-text with speaker diarization and word-level timestamp alignment.

• Features intelligent clipping logic that automatically removes silence, filler words, and detects high-impact 'golden quotes' and debate segments.

• Generates context-aware articles for platforms like blogs, Zhihu, or WeChat by synthesizing transcripts and structured insights.

• Employs a parallel execution strategy using sub-agent spawning for the clipping and extraction phases to maximize throughput and minimize processing time.

• Provides a complete workspace environment, organizing all intermediate and final assets into a structured project directory.

• Input: Requires a video file (MP4/etc.), with optional parameters for project name, output scope, target platform, and desired clip count.

• Output: A comprehensive directory containing transcript.txt, transcript.json, insights.md, clips/ (video assets), and articles/ (Markdown docs).

• Prerequisites: Requires ffmpeg, whisperX, and a HuggingFace token for speaker diarization. Supports nohup-based background processing to handle long-duration tasks reliably.

• Usage Tip: The insight-extractor phase is mandatory for generating high-quality clips and articles; skipping it may reduce the output quality of subsequent stages.