Reliability and Risk-of-Bias in the Evaluation of LLM-based Agents: A Systematic Literature Review (dataset + materials)

Open research materials for a PRISMA-2020 systematic literature review on the reliability and evaluation of LLM-based agents (51 included studies). Contains: extraction dataset (51 studies x 20 fields), adapted ROBINS-I tool, disclosure-scoring schema, PRISMA flow (SVG), checklist (2020), search/reporting extension (PRISMA-S), full-text re-rating summary, and the manuscript draft (LaTeX + HTML). All counts/citations derived from real arXiv and Semantic Scholar API responses; no fabricated data. Published as independent researcher.