synthesis-writer_skill

Overview

This skill synthesizes extracted study data into a structured, traceable synthesis saved as output/SYNTHESIS.md. It turns papers/extraction_table.csv into a comparative narrative, highlights heterogeneity and disagreement, and explicitly documents limitations and bias. Writing is evidence-first: no claims are added without table support.

How this skill works

The skill reads papers/extraction_table.csv and composes a methods snapshot, included-studies summary, theme-based findings, and a bias+limitations section. It groups studies by extraction fields (theme, intervention, outcome, study type) and reports agreements, disagreements, and sources of heterogeneity. Where evidence is missing or ambiguous, statements are flagged as "needs more evidence" rather than asserted.

When to use it

• After screening and extraction are complete and papers/extraction_table.csv exists
• When risk-of-bias assessment fields are populated in the extraction table
• When you need a reproducible synthesis tied to extracted rows for review writing
• Before drafting final discussion/conclusions to ensure claims are evidence-backed

Best practices

• Keep synthesis claims strictly tied to extraction fields; cite row counts and descriptors for each claim
• Group studies by explicit extraction fields (theme/intervention/outcome) rather than impressions
• Report both agreement and disagreement: quantify how many studies support each finding
• Tie limitations to observed gaps (missing baselines, inconsistent outcome measures, sparse negative results) rather than generic statements
• Flag any claim that cannot be supported by extraction_table.csv as "verification needed" or move it to future work

Example use cases

• Compose a synthesis section for a systematic review after completing data extraction and RoB assessment
• Compare intervention effects across subgroups using extraction_table.csv fields (population, dose, outcome)
• Document why conclusions are limited when outcome measures or follow-up windows vary across studies
• Produce a bias-focused summary showing how RoB patterns affect confidence in each theme