Security context

What an agent needs to avoid regressing past fixes and find the next vuln in this repo.

RVC-Boss/GPT-SoVITS

              main @ 48b1a01
            

Fixes

CVEs

CRITICAL

Peak severity

0.0%

Coverage

Highlights

Command Injection: 2 prior fixes. Scrutinize any change in this area.

tools/uvr5/webui.py: most-fixed (2 issues). Treat as high-risk during review.

4 high-severity fixes in this history; regressions here are high-impact.

Recurring patterns

The bug types that recur here, drawn from past fixes, not open vulnerabilities.

Command Injection: Multiple fixes and CVEs show unsanitized user input flowing into os.system calls for audio processing, enabling arbitrary command execution. The fix pattern of quoting or replacing with libraries must be applied consistently across all audio operations.

Deserialization of Untrusted Data: Unsafe torch.load of user-controlled model paths allows arbitrary code execution via pickle. The fix set weights_only=False, but CVEs show similar unsafe loads in other components (AudioPre, bsroformer, inference_webui, process_ckpt) that remain unpatched.

Command Injection: CVE-2025-49836 shows unsanitized input in change_label function leading to command injection, and CVE-2025-49833 in open_slice. The historical fix for resource leaks (3a167888) also involved os.system, indicating this component remains a high-risk surface.