| name | paper_triage |
| description | Triage a paper for biological relevance, extract the main claims, and separate evidence-backed takeaways from abstract-only impressions. |
| category | bio/literature |
| version | 1 |
| requires_tools | ["ncbi_eutils","evidence_retrieval","evidence_review","python_repl"] |
| requires_network | true |
| user_invocable | true |
| tags | ["paper-triage","abstract","literature","relevance"] |
| aliases | ["abstract_triage","paper_relevance_check"] |
| species | any |
| modality | literature |
| stage | interpretation |
| stability | stable |
| safety_level | low |
Paper Triage
Purpose
Decide whether a paper is worth deeper follow-up and summarize the strongest claims, methods, and limitations with clear provenance.
When to use
Use this skill when the user pastes an abstract, provides a title or PMID, or asks whether a paper is relevant to perturb-seq, single-cell RNA-seq, or a related biology question.
Required inputs
- paper details: abstract text, title, DOI, PMID, or citation fragment
- research question (optional): the biological question the user wants the paper judged against
Steps
- Normalize what the user supplied: abstract-only, title-plus-question, or PMID/citation-backed request.
- If a title, citation, or PMID is available, use
ncbi_eutils or evidence_retrieval to ground the paper metadata before judging relevance.
- If the user is asking a question about what the paper supports, use
evidence_review to distinguish supported conclusions from unresolved claims whenever enough source material is available.
- Use
python_repl if helpful to organize the paper into a compact table covering relevance, methods, claims, and limitations.
- If only abstract text is available, say explicitly that the triage is abstract-only and lower confidence than a metadata-backed or evidence-reviewed pass.
- Return the triage result with provenance, caveats, and the most useful follow-up action.
Output format
- Biological context or assumptions: the disease, system, assay, or research question used to judge relevance.
- Evidence or source basis: whether the triage came from
ncbi_eutils, evidence_retrieval, evidence_review, or abstract-only text.
- Relevance and claims: High / Medium / Low fit plus the main claims and key methods.
- Caveats or ambiguity: missing full text, abstract-only limitations, or unclear alignment with the user's question.
- Recommended next step: retrieve full evidence, compare with a second paper, or move into a deeper evidence review.
Failure modes
- No usable citation details: say the triage is provisional and based only on the pasted text.
- Metadata mismatch: call out title, PMID, or abstract inconsistencies before summarizing.
- Over-broad relevance question: ask what assay, disease, or mechanism the user cares about most.
Examples
- "Is this paper relevant to Perturb-seq in T cells?"
- "Triage this abstract and tell me whether it is worth reading in full."