Installable skillMIT

Literature Search & Review

# Literature Search & Review

BY beita6969UPDATED Oct 27, 2026
INTENDED FOR✳Claude Code>_Codex◈OpenClaw↗CursorInstall in a listed agent.

instructions.md

---
name: literature
description: # Literature Search & Review
---

# Literature Search & Review

## Overview
Comprehensive academic literature search and synthesis across 15+ sources.

## Capabilities
- Multi-database parallel search (PubMed, arXiv, bioRxiv, medRxiv, OpenAlex, Semantic Scholar, Crossref, DBLP, CORE, DOAJ, Europe PMC)
- Web search via Agent-Reach (Exa semantic search, Jina Reader for any URL/PDF)
- Social academic search (Twitter/X threads, YouTube talks, GitHub repos)
- Structured literature reviews with citation networks
- Knowledge gap identification
- Hypothesis generation from literature analysis

## Search Strategy
1. **Query optimization**: Short, keyword-based queries (max 7 words). PubMed-friendly syntax.
2. **Multi-source fan-out**: Parallel queries across all sources for maximum coverage.
3. **Deduplication**: By PMID, DOI, then normalized title.
4. **Reflection loop**: Evaluate coverage → identify gaps → generate follow-up queries.

## Citation Rules
- ZERO hallucinated citations. Every reference must come from real search data.
- Unified article schema: source_type, title, authors, year, journal, url, doi, pmid, abstract.
- Numbered references [1], [2], [3] — built only from retrieved articles.

## Best Practices
- Start broad, then narrow. First search finds the landscape; follow-up queries fill gaps.
- Cross-domain search. The breakthrough paper might be in an unexpected field.
- Check preprints AND published papers. Recent findings may only be on bioRxiv/arXiv.
- Verify high-impact claims. Use Semantic Scholar's citation count to identify landmark papers.
COMPATIBILITY & REFERENCES

Platform guidance

Imported from the listed AgentSkill.sh category. Complete supplied package files and notices are retained. Required external tools and credentials are not installed. No bundled code was executed during import. Individual agent task behavior has not been tested.