Agentic video understanding

1# Agentic video understanding
2 
3Hireable understanding layer. The model takes a goal and decides what to watch, at what speed, and through which modality (frames, audio, transcript), fetching only the moments needed. Vendor claims: up to ~66% lower cost and ~88% fewer tokens vs static fixed-FPS ingest, with higher accuracy.
4 
5## What this is / is not
6 
7**Is:** goal → watch only what you need → timestamps + quotes + confidence.
8 
9**Is not:** a video editor. Do not cut, overlay, caption-burn, render, schedule, post, email, or write CRM from this skill. Hand cuts to Overlap, FFmpeg, or `net-new-video-editor`. Approvals stay with the calling lane.
10 
11## When to use
12 
13- Pre-call / sales-call mining: buyer objection, next step, competitive mention
14- Shortform scoring: find a 3-second standalone hook and in/out points
15- Longform / X research: named-person + contrast moments in podcast or YouTube tape
16- Talent review: bar evidence in a Loom or trial recording
17- Client audit: every mention of a keyword across a discovery recording
18 
19Skip when the job is already a clean transcript and you only need text search.
20 
21## Inputs
22 
23| Field | Required | Notes |
24|-------|----------|-------|
25| `source` | yes | URL or local media path the runtime can read |
26| `goal` | yes | One sentence retrieval goal |
27| `keywords` | no | Extra strings to bias retrieval |
28| `max_moments` | no | Default 5 |
29| `modality` | no | `auto` (default), `frames`, `audio`, or `transcript` |
30 
31## Process
32…