systematic-review-prisma

Systematic Review PRISMA

Use this skill when the literature review needs reproducibility beyond a normal narrative SOTA. The goal is inspectable search and screening, not bureaucratic formatting.

Read First

• references/repository-contract.md
• references/systematic-review-policy.md
• references/source-ledger.md
• references/mcp-catalog.md
• references/citation-chasing.md

Workflow

1. Define research questions, review type, inclusion/exclusion criteria, and date range before broad searching.
2. Record databases and exact query strings in sota/search-strategy.md.
3. Export raw search results to sources/metadata/.
4. Expand the candidate set with citation chasing from seeds and included works: backward references and forward citations. See references/citation-chasing.md and record rounds in sota/citation-chasing-log.csv.
5. Deduplicate by DOI, arXiv ID, PMID, title, and author/year.
6. Record screening decisions in sota/screening-decisions.csv.
7. Update counts and reasons in sota/prisma-flow.md.
8. Promote included works to sources/source-ledger.csv and sota/literature-matrix.csv.
9. Track full-text retrieval and linear reading in sota/reading-log.csv for included works that support durable claims.
10. Write synthesis in sota/synthesis.md and gaps in sota/gaps.md.

Protocol Fields

Record these before screening begins:

• review type: systematic review, scoping review, rapid review, mapping review, or narrative review with reproducible search
• research questions and subquestions
• databases and interfaces, including API/MCP/manual search date
• full Boolean/query strings and filters
• inclusion criteria, exclusion criteria, and controlled exclusion reason labels
• deduplication fields and tie-break rules
• screening stages, reviewer count or agent/human roles, and disagreement rule
• extraction fields and quality/risk-of-bias rubric
• stopping rule for search expansion

If these fields are missing, produce a protocol draft first instead of running a pretend systematic review.

Screening Rules

• Separate title/abstract screening from full-text screening.
• Keep exclusion reasons short and controlled.
• Do not silently delete duplicates; count them.
• If the review is not systematic, label it narrative or scoping.
• Never invent a PRISMA count. Unknown counts stay unknown until the raw search export or screening file exists.
• Full-text unavailable is an exclusion or limitation reason, not a license to rely on abstract-only claims.
• For AI-assisted evidence synthesis, check whether PRISMA-trAIce or RAISE-style transparency notes are relevant and record AI tool use, prompts, model/version, human oversight, and validation method.

Extraction And Synthesis

Extract each included source into sota/literature-matrix.csv with method, population/dataset, intervention/system, comparator/baseline, metric/outcome, key result, limitations, and reusable claims. Synthesis must group by themes, methods, evidence quality, contradictions, and gaps. Do not force themes when the paper set is too small or heterogeneous.

Output Contract

At minimum, maintain:

• sota/search-strategy.md
• sota/screening-decisions.csv
• sota/prisma-flow.md
• sota/literature-matrix.csv
• sota/citation-chasing-log.csv
• sota/reading-log.csv
• sota/synthesis.md
• sota/gaps.md
• wiki/log.md