AgentSkillsCN

literature-agent

在 PubMed 上搜索与基因、变异、疾病或特定主题相关的生物医学文献,并生成结构化的文献综述。

SKILL.md
--- frontmatter
name: literature-agent
description: Search PubMed for relevant biomedical literature on a gene, variant, disease, or topic and produce a structured literature summary

Search PubMed for relevant literature on: $ARGUMENTS

Use the MCP tools available to you to find, contextualize, and summarize biomedical publications. Follow the steps below in order. If a step fails or returns no data, note the gap and continue.

Data Gathering Steps

1. Identify the Topic Type

Determine whether the input is:

  • A gene symbol (e.g., BRCA1, TP53) — if so, also gather gene context in step 2
  • A variant (e.g., BRAF V600E, NM_007294.4:c.5266dupC) — if so, also gather variant context
  • A disease or general topic (e.g., Lynch syndrome, CRISPR gene therapy) — skip to step 3

2. Gene/Variant Context (if applicable)

If the topic includes a gene symbol:

  • Call datasets_summary_gene with the gene symbol (taxon: human) to get the gene's full name and summary.
  • Call uniprot_search with the gene symbol and organism_id:9606 (reviewed: true) to get the canonical protein name and function — use just the first result's protein name for context.

If the topic includes a specific variant:

  • Call clinvar_search with the variant identifier to get clinical significance and associated conditions.

This context helps you write a better introduction and interpret the literature results.

3. Literature Search — Primary Query

  • Call pubmed_search with the user's topic as the query, max_results: 15.
  • Review the returned articles: titles, journals, dates, and abstracts.

4. Literature Search — Refined Queries (if needed)

If the primary search returns fewer than 5 results, or the topic is broad, run 1-2 additional targeted searches:

  • For a gene: try "[GENE] AND review[pt]" to find review articles
  • For a variant: try "[GENE] AND [VARIANT] AND pathogenic"
  • For a disease: try "[DISEASE] AND molecular mechanism"
  • For a therapy: try "[THERAPY] AND clinical trial"

Call pubmed_search for each refined query with max_results: 10.

5. Deduplicate and Rank

Remove duplicate articles and rank by relevance:

  • Recency (prioritize last 5 years unless older landmark papers are relevant)
  • Citation count (if available through Semantic Scholar integration)
  • Journal impact
  • Directly addresses the topic

Present the top 8-12 articles in the report.

Report Format

code
# Literature Summary: [TOPIC]

## Topic Overview

### Query
User's search term: [topic]

### Context
If the topic is a gene or variant, provide:
- Gene: [symbol], Full name: [name], UniProt: [accession]
- Function: [brief description from UniProt or NCBI]
- Clinical relevance: [summary of disease associations if applicable]

Or if it's a disease/topic:
- Condition: [name]
- Key characteristics: [brief description]

## Literature Search Results

### Search Strategy
- Primary query: [what was searched]
- Refined queries (if any): [additional searches]
- Total articles found: [N] (showing top [M] most relevant)

## Key Articles

| # | Title | Authors | Journal | Year | PMID | Key Finding |
|---|-------|---------|---------|------|------|-------------|
| 1 | ... | ... | ... | ... | ... | ... |
| 2 | ... | ... | ... | ... | ... | ... |

[For each of top 5-8 articles, include:]

### Article [N]: [Short Title]
- **PMID:** [ID]
- **Authors:** [first author et al.]
- **Journal:** [journal name], Year: [year]
- **Abstract Summary:** [1-2 sentences describing the research]
- **Key Methods:** [study design, sample size if applicable]
- **Key Findings:** [main results relevant to your topic]
- **Relevance:** [why this is important for understanding your topic]

## Thematic Analysis

### Main Research Themes
- [Theme 1]: [N articles] — [brief description of what these papers focus on]
- [Theme 2]: [N articles] — [brief description]
- [Theme 3]: [N articles] — [brief description]

### Consensus Findings
[1-2 paragraphs summarizing what the literature agrees on]

### Areas of Debate or Uncertainty
[Controversies, conflicting findings, or open questions in the literature]

### Recent Advances
[What's new in the field in the last 1-2 years, if applicable]

## Recommended Reading Path

For someone new to this topic, read in this order:
1. [Review article or overview paper] — comprehensive background
2. [Landmark or high-impact paper] — foundational understanding
3. [Recent paper] — current state of the field

## Summary & Next Steps

### Key Takeaways
- [3-5 bullet points summarizing the literature]

### Questions for Further Investigation
- [Open questions or gaps in current knowledge]

### Resources
- PubMed search strategy for deeper exploration: [query to use]
- Related topics to explore: [suggestions]

## Data Sources
- PubMed: [number of results, success/failure]
- Semantic Scholar (if integrated): [citation data availability]

Important Notes

  • Be accurate. Report only what is stated in article titles and abstracts. Do not invent findings.
  • Be comprehensive. Capture the breadth of research on this topic — show themes, not just individual papers.
  • Be fair. Acknowledge both supportive and critical perspectives in the literature.
  • Be concise. Summarize in plain language suitable for a researcher who may not specialize in this area.