Skip to content

Latest commit

 

History

History
 
 

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 
 
 
 
 
 
 
 
 

README.md

Literature Deep Research Skill (Enhanced)

Comprehensive literature research with target disambiguation, evidence grading, mandatory completeness checklist, and biological model synthesis. Produces detailed reports focused on content, not methodology.

Type: Strategy Skill (for AI agents)
Version: 5.0.0
Last Updated: 2026-02-04


Key Features (v5.0)

Feature Description
Target Disambiguation Resolve IDs, synonyms, naming collisions BEFORE literature search
Evidence Grading T1-T4 tiers for every claim (mechanistic → mention)
Mandatory Completeness 15 required sections; must state "limited evidence" if empty
Collision-Aware Search High-precision seeds + negative filters for ambiguous names
Citation-Network First For sparse targets, expand via citations before broad keyword search
Biological Model Synthesize evidence into 3-5 testable hypotheses
Report-Only Output Methodology in separate appendix only if requested
Scalable Bibliography Narrative stays readable; full bibliography in JSON/CSV

When to Use This Skill

Apply when users:

  • Ask "what does the literature say about [target/topic]?"
  • Need comprehensive literature coverage on a biological target
  • Request target profiles or druggability assessments
  • Want research gaps and testable hypotheses identified
  • Need literature for grant writing with evidence grades

Workflow Overview

Phase 0: CLARIFY
  └─ Is this a biological target? What scope? Methods appendix needed?

Phase 1: TARGET DISAMBIGUATION (default ON for biological targets)
  ├─ Resolve official IDs (UniProt, Ensembl, NCBI, ChEMBL)
  ├─ Gather synonyms + identify naming collisions
  ├─ Get protein architecture, expression, pathways
  └─ Output: Target Profile + Collision-aware search plan

Phase 2: LITERATURE SEARCH (internal - not shown in report)
  ├─ High-precision seed queries (build mechanistic core)
  ├─ Citation network expansion from seeds
  ├─ Collision-filtered broader queries
  └─ Theme clustering + evidence grading

Phase 3: REPORT SYNTHESIS
  ├─ Progressive writing with mandatory sections
  ├─ Apply evidence grades (T1-T4)
  └─ Generate biological model + testable hypotheses

OUTPUT:
  - [topic]_report.md (main deliverable)
  - [topic]_bibliography.json (full papers)
  - methods_appendix.md (only if requested)

Evidence Grading System

Tier Symbol Criteria Example
T1 ★★★ Mechanistic study with direct evidence CRISPR KO + rescue
T2 ★★☆ Functional study showing role siRNA phenotype
T3 ★☆☆ Association/screen hit GWAS, DepMap
T4 ☆☆☆ Review/mention/text-mined Review article

In report:

ATP6V1A drives lysosomal acidification [★★★: PMID:12345678] and regulates
autophagy [★★☆: PMID:23456789].

Mandatory Report Sections

ALL 15 sections required - state "limited evidence" if data unavailable:

  1. Identity/Aliases - IDs, synonyms, collisions
  2. Protein Architecture - Domains, isoforms, sites
  3. Complexes/Partners - Interactors with evidence type
  4. Subcellular Localization - Locations with confidence
  5. Expression Profile - Tissues, specificity
  6. Core Mechanisms - Function with evidence grades
  7. Model Organism Evidence - KO phenotypes or "none found"
  8. Human Genetics/Variants - Constraints, ClinVar, GWAS
  9. Disease Links - With evidence strength grades
  10. Pathogens - Or "none identified"
  11. Key Assays/Readouts - Available assays
  12. Research Themes - ≥3 papers per theme
  13. Open Questions/Gaps - What's unknown
  14. Biological Model - Integrated synthesis + 3-5 hypotheses
  15. Conclusions - Confidence assessment

Output Files

File Content Generated
[topic]_report.md Main narrative report Always (default deliverable)
[topic]_bibliography.json Full deduplicated papers Always
[topic]_bibliography.csv Tabular bibliography Always
methods_appendix.md Methodology details Only if user requests

Files in This Skill

File Purpose
SKILL.md Complete strategy guide
QUICK_REFERENCE.md One-page cheat sheet
TOOL_NAMES_REFERENCE.md Database tool names
README.md Overview (this file)

Tool Categories Used

Literature Search

PubMed_search_articles, EuropePMC_search_articles (use source='PPR' for preprints), openalex_literature_search, SemanticScholar_search_papers

Citation Analysis (with fallbacks)

PubMed_get_cited_byEuropePMC_get_citations PubMed_get_relatedSemanticScholar_search_papers

Annotation (fills gaps when literature sparse)

UniProt_*, InterPro_get_protein_domains, GTEx_*, HPA_*, GO_*, Reactome_*, OpenTargets_*


What Changed in v5.0

Previous Enhanced
Literature search first Target disambiguation first (default ON)
All papers treated equally Evidence grading (T1-T4)
Optional sections Mandatory completeness checklist
Simple keyword search Collision-aware query strategy
Keyword expansion Citation-network-first for sparse targets
Embedded methodology Report-only; methodology in appendix
Inline bibliography Separate scalable JSON/CSV files
Descriptive report Biological model + testable hypotheses
Any OA check method Best-effort OA with clear labeling

Quality Checklist (Must Pass)

  • All 15 mandatory sections present (or marked "limited evidence")
  • Evidence grades applied to all major claims
  • Target identifiers resolved
  • Naming collisions documented
  • ≥3 papers per research theme (or noted as insufficient)
  • Biological model synthesized
  • ≥3 testable hypotheses with experiments
  • Bibliography files generated
  • Methodology NOT in main report

Example: What Makes a Good Report

Section with proper evidence grading:

## 6. Core Mechanisms

### 6.1 Molecular Function
ATP6V1A is the catalytic A subunit of the V-ATPase complex, responsible for 
ATP hydrolysis that drives proton translocation [★★★: PMID:12345678].

**Evidence Quality**: Strong (5 mechanistic, 2 functional studies)

### 6.2 Biological Role  
V-ATPase acidification of lysosomes is essential for:
- Autophagosome-lysosome fusion [★★★: PMID:23456789]
- mTORC1 activation via amino acid sensing [★★☆: PMID:34567890]
- Bone resorption by osteoclasts [★★★: PMID:45678901]

Testable hypothesis:

| # | Hypothesis | Perturbation | Readout | Expected | Priority |
|---|------------|--------------|---------|----------|----------|
| 1 | ATP6V1A loss impairs autophagy | siRNA knockdown | LC3 puncta, p62 levels | ↑LC3-II, ↑p62 | HIGH |
| 2 | V-ATPase inhibition blocks mTORC1 | Bafilomycin A1 | S6K phosphorylation | ↓pS6K | HIGH |

Getting Started

  1. Read SKILL.md for complete strategy
  2. Use QUICK_REFERENCE.md during execution
  3. Check TOOL_NAMES_REFERENCE.md for exact tool names

Version 5.0.0 - Enhanced with target disambiguation, evidence grading, and biological model synthesis