| name | gaussian-youtube-video-wall-evolver |
| description | Use this skill when you need to continuously collect the latest YouTube research/demo videos for realtime Gaussian avatar systems, publish a curated video-wall dataset, and keep a self-evolving memory of queries/channels over repeated cycles. |
Gaussian YouTube Video Wall Evolver
Overview
Run repeatable discovery cycles for realtime Gaussian avatar demos on YouTube, then publish a curated wall dataset used by both slides and website pages.
This skill:
- Queries YouTube search result pages for Gaussian-avatar demo terms.
- Extracts video entries from
ytInitialData.
- Filters to realtime avatar-relevant demos.
- Updates a self-evolving state with better focus terms and channel priors.
- Publishes data for UI embedding and modal playback.
Run Cycle
- Run one cycle:
python .claude/skills/gaussian-youtube-video-wall-evolver/scripts/evolve_gaussian_video_wall.py
- Useful options:
--max-videos 24
--max-per-query 18
--no-publish to keep outputs local only
Outputs
Local skill outputs:
.claude/skills/gaussian-youtube-video-wall-evolver/references/latest-videos.json
.claude/skills/gaussian-youtube-video-wall-evolver/references/latest-report.md
.claude/skills/gaussian-youtube-video-wall-evolver/references/progress.md
.claude/skills/gaussian-youtube-video-wall-evolver/references/history/videos-*.json
.claude/skills/gaussian-youtube-video-wall-evolver/state.json
.claude/skills/gaussian-youtube-video-wall-evolver/events.jsonl
Published mirrors (unless --no-publish):
web/app/data/gaussian-video-wall.json
web/public/docs/gaussian-video-wall-latest.json
web/public/docs/gaussian-video-wall-latest.md
Self-Evolving Contract
Each cycle must:
- Load previous
state.json and latest-videos.json if present.
- Compute and persist delta of newly surfaced videos.
- Derive new focus terms from fresh titles/channels.
- Persist focus terms and channel priors to
state.json.
- Append cycle record to
events.jsonl and progress.md.
Do not delete history/events unless explicitly requested.