video-datasetlisted
Install: claude install-skill vidnavigator/skills
# Build a dataset from videos
Find the videos, extract the same fields from each one, write one row per video. The fields come from what is said in each video and, with `include_scenes`, from what it shows, read by `extract_video_data`: a search result's caption is not a source for them.
## Calling VidNavigator
Each step names a VidNavigator tool and its arguments. Call it in whichever way this session offers:
- **MCP tools.** If tools such as `extract_video_data` are in your tool list (their names may carry a prefix, for example `mcp__vidnavigator__extract_video_data`), call them directly. A tool that starts a background task returns a `task_id` and a `task_type`: call `get_task_status` with both until `task_status` is `completed` or `failed`. That call waits for the task, so make it again whenever it still answers `processing`. TikTok results come 50 per page: pass `pagination.next_cursor` as `cursor` for the next page.
- **The bundled CLI.** Otherwise run the script that sits beside this file, where `SKILL_DIR` is the directory containing this SKILL.md:
```bash
node "$SKILL_DIR/scripts/vn.mjs" extract_video_data video_url="https://www.tiktok.com/@user/video/123" schema=@schema.json --out rows/123.json
```
Same tools, same arguments, written `key=value`. Quote a value that holds spaces, `&` or `?`. An object such as `schema` is read from a file: `schema=@schema.json`. The CLI waits for background tasks and gathers every page of a search. `--out FILE` writes th