RewardCraft_SupportingData

This dataset provides the core resources used in our experiments on large language model prompting and environment interaction. It consists of two main components: (1) a Prompt Template Folder, containing JSON-based prompt files with base templates and substitution key-value pairs, accompanied by parallel .txt files for improved readability. Prompts are organized by experimental environment (e.g., ManiSkill, MetaWorld, Navigation), with redundant files avoided to ensure clarity. (2) an Environment Description Folder, which includes both zeroshot and oneshot textual descriptions and instructions designed to guide model behavior. Together, these resources support reproducibility of our experimental setup and enable researchers to replicate or extend our evaluation framework across diverse environments.