| name | deep-exploration |
| description | Saves time by autonomously reading transcripts, synthesizing arguments, and generating direct jump-links to key moments. |
Skill: Deep Exploration & Timestamp Targeting
Objective: Save the user from watching hours of video by autonomously reading transcripts, synthesizing the core arguments, and generating direct jump-links to the exact moments of interest.
Execution Steps:
- Locate: Use
search_youtube to find 3 to 5 highly relevant videos on the specific query. Prioritize videos from different creators to get diverse perspectives.
- Ingest (Read): For each video, use
get_video_transcript(video_id) to pull the full closed captions with timestamps.
- Fallback: If the transcript is unavailable, fall back to using
get_video_details to read the video's description instead.
- Cross-Reference & Target: Analyze the transcripts to find the exact moments where the creator answers the user's specific query.
- Link Generation (MANDATORY): For the most valuable moments (clips), you MUST use
generate_timestamp_url(video_id, timestamp) to create a direct clickable URL. Do NOT try to guess the ?t= parameter yourself.
- Deliver: Present a highly curated "Knowledge Report". For each video, provide a 2-sentence summary of their unique angle, followed by the direct clickable URL to the exact timestamp where the key insight begins.
HTML Publishing Rule: If the user asks to publish this report to HTML, you MUST ensure that the <a> href attributes in your HTML string use the exact URLs returned by the generate_timestamp_url tool. Never use plain https://youtu.be/ID when a specific moment is discussed.
Next Actions: Ask the user if they want to publish this curated list of clips into a permanent HTML Knowledge Report, or if they have any feedback on the tool's accuracy.
