Search

What this skill does

The Search skill enables precise retrieval of specific moments within videos by querying indexed spoken content or visual scenes. It supports natural language queries, exact keyword matching, and scene-based search powered by AI-generated descriptions, allowing marketers to locate relevant video segments without manual review. Videos must be indexed first, either by transcribed speech or scene extraction, to enable efficient, accurate search operations.

Who it's for

This skill is designed for performance marketers curating video ads or educational content who need to quickly locate specific messaging or product mentions. Growth leads managing video assets across platforms like YouTube or TikTok can use it to optimize content reuse and targeting by extracting high-impact moments. SEO and content strategists working with video transcripts benefit from semantic and keyword search to align video content with audience queries and improve discoverability.

Key workflows

First, practitioners index videos by transcribing spoken words or generating scene descriptions based on shot or time-based segmentation. Next, they run semantic or keyword searches on the spoken transcription to find segments matching natural language queries or exact phrases. For visual content, they perform semantic search against the scene index, optionally filtering by custom metadata like camera angle or action type. Finally, users review results by extracting individual clips, compiling matched segments into highlight reels, or streaming selected shots for further analysis or distribution.

Common questions

Can I search a video without indexing first? No, indexing spoken words or scenes is a mandatory one-time setup per video before any search queries will work. How do I handle multiple scene indexes on a single video? You can create and target specific scene indexes using the returned scene_index_id to isolate searches per index. Is semantic search better than keyword for videos? Semantic search captures intent and context in queries, making it more effective for finding relevant spoken or visual content beyond exact term matches.

How to use in Metaflow

Attach the Search skill to your Metaflow agent task by linking it to a video that has been indexed for spoken words or scenes. When invoked, the agent executes natural language or keyword searches and returns time-stamped video segments matching your query, ready for streaming or export. This skill fits seamlessly into pipelines requiring granular video content extraction and analysis, enabling rapid insight generation from large video libraries.

For broader context, see our roundup of marketing skills claude, and read Claude skills for SEO for related setup guidance.