LLM Security and Temporal Analysis Data

The dataset is derived from processed tables built from original experimental runs containing generated dependency manifests, prompts, raw LLM outputs, and run metadata.It contains:- 521,539 rows and 86 columns- 31,648 compliant manifest-generation trials- 8 LLMs from OpenAI, Anthropic, xAI, and Meta- 10 application scenarios- 5 languages/ecosystems- 6 manifest formats- 4 prompting strategies- 4,078 distinct registry-normalized packagesThe code, runs, and per-run tables are stored in LLM-Package-Security/. The combined dataset is available as all_runs_with_osv.csv, and analysis.ipynb reproduces the study’s results.The dataset is intended to support the analysis of the security and temporal implications of LLM-generated package recommendations.