| name | literature-search |
| description | Run comprehensive literature search for Research Harness. Trigger on phrases like "literature-search", "/literature-search", "文献检索", "系统检索论文", "帮我搜相关论文", or equivalent requests to search, shortlist, and ingest relevant papers. |
| public_suite | research |
| task_type | discovery |
| data_access_level | local+external |
| inputs | topic scope, time range, venue/domain constraints, topic_id |
| outputs | ranked paper candidates, ingestion decisions, frontier evidence |
| gates | ingestion requires topic_id; frontier claims need academic and industry signals |
| artifact_schemas | paper_pool, frontier_attention_evidence |
| estimated_runtime | 10-45 min |
Literature Search
Use this skill for broad paper discovery.
Workflow
- Read
~/code/research-harness/docs/agent-guide.md if topic context is unclear.
- Clarify or infer:
- research topic
- time range
- venue or domain constraints
- Use
paper_search or rhub-compatible search flow.
- Deduplicate and rank results by relevance.
- Recommend which papers to ingest and ingest them when the user intent is clearly operational rather than exploratory.
Output
- concise search query formulation
- top relevant papers
- ingestion decision or next-step recommendation
Frontier Attention Evidence
When the search supports a paper, survey, or topic-validation task, include a
frontier-attention pass:
- latest top academic signals: recent top-venue papers, best/oral/spotlight
papers, benchmarks, surveys, and workshop/tutorial activity
- industry signals: official lab/company blogs, technical reports, SDKs,
standards, open-source frameworks, and product releases
- trend summary: why the topic is timely now, and which scope boundaries the
evidence does or does not support
Papers must be ingested under the topic; non-paper sources should be recorded
in a frontier_attention_evidence artifact with URL, date, source type, and
the claim they support.
