zaakirio/

coding_harmless_prompts

Benign coding and technical prompts for the harmless side of infosec refusal-direction extraction.

This dataset contains benign coding and technical prompts intended to be paired with infosec_harmful_behaviors. The contrast helps isolate malicious coding intent rather than a general coding or technical-domain direction. Rows:

train: 400 test: 120

Schema:

text: prompt string

Use this dataset… See the full description on the dataset page: https://huggingface.co/datasets/zaakirio/coding_harmless_prompts.