Securing IaC Datasets

This repository contains two datasets used for research on securing Infrastructure as Code (IaC), specifically targeting Terraform code smells.

1. COMPSAC 2025 Dataset (compsac25/)

This dataset was developed for paper Harnessing the Power of LLMs for Code Smell Detection in Terraform Infrastructure as Code at IEEE COMPSAC 2025. It consists of a synthetic collection of Terraform files categorized by various code smells.

Location: compsac25/synthetic_dataset/

Categories included:

• admin_by_default
• complex_expression
• deprecated_statement_usage
• empty_password
• hard-coded_secret
• incomplete_conditional
• incomplete_tasks
• insufficient_key_size
• invalid_port_ranges
• invalid_property_value
• law_of_demeter
• suspicious_comment
• unrestricted_ip_address
• weak_crypto__algo_

Publication & Citation If you use this dataset or our methodology, please cite our IEEE COMPSAC 2025 paper:

@inproceedings{vo2025harnessing,
  title={Harnessing the Power of LLMs for Code Smell Detection in Terraform Infrastructure as Code},
  author={Vo, Quoc-Huy and Dao, Ha and Fukuda, Kensuke},
  booktitle={2025 IEEE 49th Annual Computers, Software, and Applications Conference (COMPSAC)},
  pages={533--542},
  year={2025},
  organization={IEEE}
}

2. COMPSAC Special Issue 2026 Dataset (compsac_special_issue26/)

This dataset is an expanded version developed for our journal paper (IPSJ COMPSAC Special Issue 2026). It includes both raw Terraform files and a structured JSONL file for fine-tuning LLMs tasks.

Paper title: "Securing IaC: Detecting Terraform Security Smells with fine-tuned LLMs."

Location: compsac_special_issue26/dataset/

Contents:

• 1.code_smell_tf_file/: Raw .tf samples organized by type (admin_by_default, empty_password, hard_coded_secret, insufficient_key_size, suspicious_comment, unrestricted_ip_address, weak_crypto__algo_).
• dataset.jsonl: A structured dataset containing 1,382 entries with instructions, Terraform code snippets, and ground truth labels for code smell detection.
• dataset_creation.py: The script used to generate the dataset.

License

MIT License, see LICENSE.