Skip to content

alestronati/Systematic-Review

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

1 Commit
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Systematic Review Agent for Claude Code

A PRISMA 2020 systematic review and scoping review agent for Claude Code CLI. Covers the full review lifecycle: protocol development, search strategy, screening, data extraction, risk of bias assessment, evidence synthesis, and PRISMA-compliant reporting.

Architecture

The agent follows a 1 agent + 4 forked sub-skills architecture. The orchestrating agent delegates each review phase to a specialized skill running in isolated (context: fork) context:

systematic-review-agent (orchestrator)
  |
  |-- sr-protocol-builder        PICO/PCC, eligibility, search strategy, directory init
  |-- sr-search-and-prompt       PubMed/Scholar search, AI prompt generation, ingestion
  |-- sr-screen-and-extract      Screening decisions, data extraction, RoB assessment
  |-- sr-synthesize-and-report   Evidence tables, synthesis, PRISMA compliance, final report, audit

Each sub-skill receives a focused task, runs in its own context window, and returns results to the orchestrator. This keeps each phase within token limits while preserving methodological coherence across the full review.

Prerequisites

  • Claude Code CLI: npm install -g @anthropic-ai/claude-code

Installation

# Copy agent definition
cp agents/systematic-review-agent.md ~/.claude/agents/

# Copy skills
cp -r skills/systematic-review-agent ~/.claude/skills/
cp -r skills/sr-protocol-builder ~/.claude/skills/
cp -r skills/sr-search-and-prompt ~/.claude/skills/
cp -r skills/sr-screen-and-extract ~/.claude/skills/
cp -r skills/sr-synthesize-and-report ~/.claude/skills/

Optional Dependencies

The agent integrates with the following optional skills when available:

Skill Purpose
chain-of-verification Mandatory verification of all synthesis claims (--thorough mode)
scientific-skills PubMed API (pubmed-database), ClinicalTrials.gov (clinicaltrials-database), citation verification (literature-review), PDF generation
fetch-papers Open-access PDF retrieval via Unpaywall (free)

Quick Start

/review init [topic]       Initialize protocol and directory structure
/review search             Execute searches and generate AI prompts
/review screen             Run screening loop (title/abstract + full-text)
/review extract            Run data extraction loop
/review synthesize         Build evidence tables and final report
/review audit [path]       Audit an existing review for PRISMA 2020 compliance
/review status             Show PRISMA compliance and progress
/review checklist          Display current PRISMA tracker

Features

PRISMA Compliance

  • PRISMA 2020 (27-item) and PRISMA-ScR (22-item) compliance tracking updated after every phase
  • AI transparency reporting per PRISMA 2020 Items 8/9/11 and the emerging PRISMA-AI extension
  • SWiM (Synthesis Without Meta-Analysis) 9-item checklist integration for narrative synthesis

Reference Documentation

14 reference documents bundled under skills/systematic-review-agent/references/:

  • PRISMA 2020 checklist and Explanation & Elaboration
  • PRISMA-ScR checklist with all 22 TipSheets
  • PRISMA-P protocol checklist (17 items / 26 sub-items)
  • PRISMA flow diagram guide (35-field data model, 4 variants)
  • SWiM guideline for synthesis without meta-analysis
  • Risk of bias tools (RoB 2, ROBINS-I, ROBINS-E, NOS, AMSTAR 2, QUADAS-2, PROBAST, QUIPS, and others)
  • Search strategy guide (PubMed MeSH, Google Scholar, Embase, PRISMA-S, MECIR standards)
  • GRADE certainty assessment (5 downgrade + 3 upgrade domains, SoF tables)
  • Biomechanical review guide (BIOMECH_CHECKLIST, FE_CHECKLIST)
  • AI prompt templates (Perplexity Pro, Elicit Pro, Gemini, Claude.ai)
  • AI transparency requirements and disclosure framework
  • 8 methodological decision trees (synthesis method, MA model, RoB tool, GRADE, reporting bias, heterogeneity, protocol deviations, review type selection)

Project Templates

9 project templates under skills/systematic-review-agent/assets/templates/:

  • Review protocol (REVIEW_PROTOCOL_TEMPLATE.md)
  • Review plan / paper checklist (REVIEW_PLAN_TEMPLATE.md)
  • Data extraction form (DATA_EXTRACTION_FORM_TEMPLATE.md)
  • Evidence table (EVIDENCE_TABLE_TEMPLATE.md)
  • PRISMA flow diagram (PRISMA_FLOW_TEMPLATE.md)
  • Screening log (SCREENING_LOG_TEMPLATE.md)
  • AI usage log (AI_USAGE_LOG_TEMPLATE.md)
  • Final report / manuscript (FINAL_REPORT_TEMPLATE.md)
  • Review audit / compliance assessment (REVIEW_AUDIT_TEMPLATE.md)

Review Automation

  • Automated screening loop (screening-loop.sh) for iterative paper-by-paper screening and extraction with fresh context per iteration
  • Deduplication by DOI/PMID across multiple search sources
  • AI prompt generation for Perplexity Pro, Elicit Pro, Gemini, and Claude.ai web UIs
  • Open-access PDF retrieval via Unpaywall integration

Supported Review Types

  • Systematic reviews (intervention effectiveness)
  • Scoping reviews
  • Diagnostic test accuracy (DTA) reviews
  • Prognostic factor reviews
  • Prevalence/incidence reviews
  • Prediction model reviews
  • Etiological/risk factor reviews
  • Biomechanical evidence reviews (cadaveric, finite element, animal model)
  • Qualitative evidence synthesis (QES)
  • Network meta-analysis (NMA)
  • Umbrella reviews
  • Living systematic reviews
  • Rapid reviews
  • Mixed methods reviews

Methodological Decision Support

8 structured decision trees for key methodological choice points:

  1. Synthesis method selection (meta-analysis vs. SWiM alternatives)
  2. Meta-analysis model selection (fixed-effect vs. random-effects)
  3. Risk of bias tool selection (by study design)
  4. GRADE certainty assessment (downgrading/upgrading)
  5. Reporting bias assessment (funnel plot, qualitative)
  6. Heterogeneity assessment (I-squared interpretation)
  7. Protocol deviation tracking
  8. Review type selection (SR vs. ScR vs. QES vs. umbrella vs. rapid)

Biomechanical Evidence

  • BIOMECH_CHECKLIST (6 domains) for cadaveric and mechanical testing studies
  • FE_CHECKLIST (6 domains) for finite element modeling studies
  • Adapted GRADE starting at Moderate for validated biomechanical studies
  • Mixed-study synthesis guidance for combining clinical and engineering evidence

Reference Documentation

All reference files are indexed in skills/systematic-review-agent/references/INDEX.md. Key lookup paths:

I need to... Reference file
Track PRISMA compliance (SR) prisma-2020-checklist.md
Track PRISMA compliance (ScR) prisma-scr-checklist.md
Get item-by-item PRISMA guidance prisma-ee-elaborations.md
Write a review protocol prisma-p-checklist.md
Choose a risk of bias tool risk-of-bias-tools.md or decision-trees.md
Build database search strings search-strategy-guide.md
Rate certainty of evidence quality-assessment-scales.md or decision-trees.md
Decide synthesis method decision-trees.md and swim-guideline.md
Generate PRISMA flow diagram prisma-flow-diagram-guide.md
Assess biomechanical studies biomechanical-review-guide.md
Report AI tool use prisma-ai-transparency.md

Attribution

This agent's methodology is informed by open-access and published standards including the Cochrane Handbook, PRISMA 2020, GRADE, and others. See ATTRIBUTION.md for the full bibliography.

License

This project is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0) license.

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages