Fix duplicate logs and add session download capability

harden Red exploit report normalization

align harness wording with Vyber branding

tighten sandbox trace accuracy

allow Vyber harness sandbox tools

stabilize Vyber harness deployment

route blue remediation through Vyber harness

verify patches through red re-attack

avoid repeated red-blue retry loops

stabilize operation terminal and agent validation

polish Vyber cyber-range experience

feat: loop-until-zero vuln engine — 3 vulns/scenario, multi-round Red+Blue agents, per-vuln scoreboard

feat: terminal-aesthetic output — timestamps, box headers, THINK/EXEC/STDOUT blocks, tight AI prompts

revert: restore correct 7B GGUF filename

fix: update 7B GGUF filename to Q4_K_M suffix

Parameterize model size options and add 7B/1.5B tiered fallback search in backend

Design update matching QuantSafe, relative paths in readme, and Modal build compile fix

fix: Add clang to apt_install to compile llama-cpp-python in container

feat: Integrate llama.cpp runtime and GGUF model downloader to achieve merit badges

fix: Switch Modal base image to nvidia/cuda-devel and disable flashinfer to resolve JIT compile errors

fix: Add vllm to container image and forward client OPENAI_API_KEY to Modal

feat: Enhance Vyber run fallback to allow direct shell execution

chore: Rename OpenCode mentions to Vyber and add CLI symlink

Implement live LLM agent loop and add interface cold-start warning

Unify scenarios.py and backend.py to fix packaging imports

Replace deprecated modal.Mount with image.add_local_file

Add scenarios.py file mount to Modal function

Fix scenarios import packaging bug

Enable live Qwen LLM model execution

Update backend.py with new structured logs layout

Deploy Vyber Cyber-Range