| name | openreview-latest-topic-search |
| description | Use this skill when you need structured paper recommendations for a topic across NeurIPS, ICML, ICLR, and CVPR. It calls the local openreview_search API, automatically uses each venue's latest locally indexed year, enables LLM relevance scoring, returns 10 papers per venue, and does not request bilingual translation. |
OpenReview Latest Topic Search
Use this skill when the user wants topic-based paper recommendations from the four main conferences and needs machine-readable JSON for downstream agent use.
Workflow
- Make sure the local
openreview_search backend is running.
- Run the bundled script. This request may take several minutes because the backend performs keyword extraction, multi-venue retrieval, and LLM relevance scoring. During testing, do not assume the agent is stuck just because there is no immediate output; allow up to 10 minutes before treating it as a timeout or failure.
python3 scripts/search_latest_topic.py --topic "your topic here"
- Return the script's JSON output directly unless the user explicitly asks for extra narration.
Behavior
- Fixed venues:
NeurIPS, ICML, ICLR, CVPR
- Each venue uses the latest year that is already indexed locally
- Each venue returns up to 10 papers
- Relevance scoring is enabled
- Title/abstract translation is disabled
- Partial failures are preserved in the JSON response under
failures
Configuration
- Default API base URL:
http://127.0.0.1:8000
- Override with env var:
OPENREVIEW_SEARCH_BASE_URL
Notes
- If the backend is unreachable or a venue has no indexed local year, fail explicitly instead of inventing results.
- This skill can be slow in real runs. Agents should treat it as a potentially long-running call and avoid declaring it hung too early when no intermediate output is shown.
- Keep the output structured. Do not rewrite field names unless the user asks for a different format.