| name | conference-papers-fetch |
| description | Use when the user asks to fetch, crawl, collect, scrape, or refresh conference paper candidates from CVPR/CVF pages, RSS/Atom feeds, manual paper lists, or topic filters before reading or building the site. |
Conference Papers Fetch
Use this thin entrypoint for the fetch stage of the conference paper website. The implementation lives in ../conference-papers; do not duplicate scripts here.
Workflow
- Resolve the site root and ensure it has or will have
data/.
- Read defaults from
../conference-papers/config/user-config.json; allow a site-local config/user-config.local.json to override conference, year, topics, source URL, and paths.
- Read the core source rules if needed:
../conference-papers/references/source-adapters.md.
- For a conference page, RSS/Atom feed, or manual JSON list, run one of:
python3 ../conference-papers/scripts/add_papers.py --draft
python3 ../conference-papers/scripts/add_papers.py --conference CVPR --year 2026 --topic VLA --url "https://cvpr.thecvf.com/virtual/2026/papers.html" --data data --draft
python3 ../conference-papers/scripts/crawl_conference.py --conference CVPR --year 2026 --url "https://cvpr.thecvf.com/virtual/2026/papers.html" --topics VLA Humanoid locomotion --out data/conferences/cvpr-2026.json
- When the conference page gives weak metadata or misses PDFs, run arXiv enrichment:
python3 ../conference-papers/scripts/enrich_arxiv.py --data data
- If the user provides several topics, preserve multi-topic membership. A paper can belong to more than one topic.
- Stop after fetching unless the user also asks to read notes or build the site.
Output
Fetch should update data/papers/*.json and data/conferences/*.json. Report how many papers were added or merged.
