Reproducibility Package for "Attack Success and Utility Cost of Defenses Against Indirect Prompt Injection in Tool-Using Large Language Model Agents: A Controlled Empirical Evaluation"

Reproducibility package for the manuscript. Contains the corrected canonical results CSV, the GPT-5.4-mini re-run per-instance traces, the McNemar recomputation output, and the analysis code. Version 2: the GPT-5.4-mini results were corrected after identifying a summary-aggregation (LOGDIR) bug in the original run; all counts in this version were verified directly from per-instance traces.