Skip to content

SECURITY.md

Latest commit

 

History

History
77 lines (58 loc) · 3.02 KB

SECURITY.md

File metadata and controls

77 lines (58 loc) · 3.02 KB
Raw

Security Policy

Supported Versions

SpatialCrafter is a research codebase. Security fixes are provided for the latest release on the default branch only.

Version Supported
1.0.x ✅
< 1.0 ❌

Reporting a Vulnerability

Please do not report security issues through public GitHub issues, pull requests, discussions, or social media.

Use GitHub's private vulnerability reporting instead:

  1. Go to the Security tab of the repository.
  2. Click Report a vulnerability to open a private advisory.

If you cannot use GitHub advisories, contact the maintainers at cfangac@connect.ust.hk .

Please include:

  • The affected version or commit hash.
  • A description of the issue and its potential impact.
  • Steps to reproduce, ideally with a minimal proof of concept.
  • Any suggested fix or mitigation.

What to expect

  • Initial acknowledgement within 5 business days.
  • An assessment and, where applicable, a remediation plan communicated after triage.
  • Credit in the release notes for the fix, unless you ask to remain anonymous.

Scope

In scope are vulnerabilities in this repository's own code, in particular:

  • Unsafe deserialization of model weights or other checkpoint files.
  • Path traversal or arbitrary file read/write reachable from configuration files, CLI arguments, or dataset split CSVs.
  • Injection, SSRF, or arbitrary code execution through remote model downloads (huggingface-cli, modelscope, or URLs passed to loaders).
  • Command injection through subprocess invocations.
  • Exposure of the optional Gradio demo service. The demo binds to 127.0.0.1 by default; if you expose it on a network interface, put it behind your own authentication and TLS termination.

Out of Scope

  • Vulnerabilities in third-party dependencies and vendored upstream projects, including TRELLIS, DiffSynth-Studio, Wan2.1 / WanVideo, Depth Anything V3, nvdiffrast, diffoctreerast, and the RaDe-GS rasterizer. Please report those upstream.
  • Loading checkpoints from untrusted sources. Model weights are deserialized with PyTorch; several code paths use weights_only=False and can therefore execute arbitrary code embedded in the file. Only load checkpoints you trust. This is documented behaviour, not a vulnerability.
  • Denial of service through resource exhaustion, such as oversized inputs that exhaust GPU memory. Inference cost scales with input resolution and frame count by design.
  • Issues that require an attacker to already control the host, the Python environment, or files the user voluntarily supplies.
  • The absence of authentication, authorization, or multi-tenancy. This project is not designed to be run as a shared public service.

Data and Privacy

The repository ships no user data collection, telemetry, or phone-home behaviour. Inference reads only the inputs you provide, and model weights are downloaded from the sources named in the README.