| name | realtime-gaussian-research-evolver |
| description | Use this skill when you need deep, repeatable research on realtime Gaussian avatars, continuous ArXiv/GitHub tracking, and auto-updated conclusions in deep-research-Realtime-Gaussian Avatars.md. |
Realtime Gaussian Research Evolver
Overview
Run repeatable research cycles for realtime Gaussian avatar systems.
Each cycle:
- pulls current evidence from ArXiv and GitHub,
- tracks deltas across cycles,
- updates a canonical "how it works" conclusion,
- persists state/events/history for self-evolution.
The primary authored output target is:
deep-research-Realtime-Gaussian Avatars.md
Run Cycle
- Run one cycle:
python .claude/skills/realtime-gaussian-research-evolver/scripts/evolve_realtime_gaussian_research.py
- Useful options:
--max-arxiv-per-query 12--max-github-search 12--max-papers 40--no-publish (skip publishing to web/public/docs)
Outputs
Local skill outputs:
.claude/skills/realtime-gaussian-research-evolver/references/latest-research.json.claude/skills/realtime-gaussian-research-evolver/references/latest-report.md.claude/skills/realtime-gaussian-research-evolver/references/progress.md.claude/skills/realtime-gaussian-research-evolver/references/history/research-*.json.claude/skills/realtime-gaussian-research-evolver/state.json.claude/skills/realtime-gaussian-research-evolver/events.jsonl
Published mirrors (unless --no-publish):
web/public/docs/realtime-gaussian-research-latest.jsonweb/public/docs/realtime-gaussian-research-latest.md
Primary narrative output:
deep-research-Realtime-Gaussian Avatars.md (upserted conclusion section)
Self-Evolving Contract
Each cycle must:
- load
state.json and prior latest-research.json when present,
- compute newly discovered papers/repos since the previous cycle,
- derive updated focus terms from new evidence,
- persist updates in
state.json,
- append an event line to
events.jsonl,
- append a human-readable run log in
references/progress.md.
Do not delete history/events unless explicitly requested.
