Feat (Phase 4): Add GitHub Action, GitLab CI template, and VS Code task

Feat (Phase 3): Implement Git hook integration and pre-commit framework support

Feat (Phase 1 & 2): Extract scanner module and add CLI interface

Deploy Option A: Transition HF Space to lightweight OpenEnv Server

Improve: Broaden CWE families for smoother reward signals

Fix: Switch to official unsloth/unsloth base image for maximum stability

Auto-skip wandb when WANDB_API_KEY is not set

Add reward tuning, improved prompt, eval harness, and serving Dockerfile

Fix: upgrade base image to PyTorch 2.6.0 for torch.int1 / torchao compat

Fix: stop pinning transformers/peft/accelerate over Unsloth's versions

Add selfcheck cache entry for Python dependencies with last check timestamp

Fix TRL version conflict: remove trl==0.12.1 pin that broke GRPOConfig import

Fix: Enable unbuffered logging for real-time training monitoring

Fix: Update Unsloth installation and improve path handling in training script

Fix: Explicit HF token handling and final GRPO batch alignment

Fix: Add libaio-dev and reorder pip installs for Unsloth stability

Fix: Pin stable versions to resolve 2026 experimental package conflicts

Fix P1/P2: Batch config, prompt synchronization, and evaluation parser consistency

Remove exclude_list.txt as it is no longer needed

Add scripts for hero case finding, hero details retrieval, and training log plotting

Improve: More robust case-insensitive XML parsing for small models

Fix: Upgrade to PyTorch 2.5.1 to resolve compatibility issues

Fix: Use robust PyTorch base image in Dockerfile

Add smoke test for random episodes and initial simulated rewards data

Initial clean deploy commit