Axon InfraLinter

Research prototype for Axon InfraLinter, an LLM-based tool to detect security misconfigurations in Terraform Infrastructure-as-Code and compare them against rule-based scanners (Checkov, tfsec).

Project layout

• requirements.txt – Python dependencies.
• axon_infralinter/
  • config.py – central configuration (paths, GitHub, model names).
  • data/github_scraper.py – discover and clone GitHub repos with Terraform.
  • data/terraform_collector.py – extract .tf files and metadata.
  • scanning/scanner.py – run Checkov/tfsec and generate unified JSON labels.
  • dataset/build_dataset.py – build balanced train/val/test splits (JSONL).
  • models/train_lora.py – fine-tune Code Llama with LoRA on the dataset.
  • models/evaluate.py – compare baselines and LLM on the held-out test set.
  • cli/axon_infralinter_cli.py – simple CLI wrapper for the trained model.

Basic usage

1. Create and activate a Python virtualenv.

2. Install dependencies:

   pip install -r requirements.txt

3. Configure axon_infralinter/config.py (output root, GitHub token, model path).

4. Run the pipeline step by step:

   # 1) Repo discovery + cloning + .tf collection
   python -m axon_infralinter.data.github_scraper
   python -m axon_infralinter.data.terraform_collector
   
   # 2) Scanning and label generation
   python -m axon_infralinter.scanning.scanner
   
   # 3) Dataset building
   python -m axon_infralinter.dataset.build_dataset
   
   # 4) LLM training and evaluation
   python -m axon_infralinter.models.train_lora
   python -m axon_infralinter.models.evaluate
   
   # 5) CLI inference on new files
   python -m axon_infralinter.cli.axon_infralinter_cli path/to/file.tf

This is a research prototype, not a production security scanner.