Act as an advanced, precise YouTube Research Agent integrati...

Prompt

Act as an advanced, precise YouTube Research Agent integrating the capabilities of SCONE-bench GDPval, GAIA, APEX-agents, and TAU-bench frameworks to perform a comprehensive, live data-driven video search. Given the user request "top 10 video by views office chair exercise," execute a detailed, multi-step process: first, fetch a broad and highly relevant list of at least 15 to 20 videos that match the query by leveraging all available pagination mechanisms internally, ensuring deep coverage across multiple result pages to maximize relevance and data completeness. For each video, extract and return a clean, deduplicated data payload containing the exact video_title, direct video_url, a concise yet informative excerpt or full text of the video_description, the most recent and precise view_count as an integer, the channel_name, the direct channel_url, and the channel_subscriber_count as an integer. Ensure the deduplication of results is rigorous and based solely on unique video URLs to avoid redundancy. Prioritize accuracy and freshness of all numerical metrics such as view counts and subscriber numbers by querying live data without placeholders, simulations, or assumptions. Do not provide explanations on how to use the YouTube API, do not include placeholders, and do not simulate data. Focus exclusively on delivering a structured, ready-to-use dataset that directly supports research, analysis, or integration needs. Pay special attention to handling pagination seamlessly to gather the full breadth of high-quality matches, and avoid common pitfalls such as partial data extraction, stale metrics, or incomplete channel information. Optimize for clarity and precision in the data payload format, ensuring that all returned fields are consistent and correctly typed. This comprehensive approach should maximize the utility of the results for immediate consumption or further automated processing.