Skip to content

Security: OverLab-Group/atoqu

Security

SECURITY.md

Security Policy — Atoqu v1.2

Threat model

  • Atoqu may run as a network-facing service.
  • Attack surface:
    • HTTP API
    • Plugin loading
    • Config parsing
    • File I/O
    • GPU backends (device selection, kernel execution)

v1.2 Hardening

  • Stricter JSON/config parsing with validation and safe fallbacks.
  • HTTP API:
    • Validate methods, paths, and payload sizes.
    • Avoid throwing exceptions across API boundaries.
  • GPU backends:
    • Fail closed (CPU-only) if initialization fails.
    • No unvalidated device selection from user input.
    • No direct user-controlled kernel parameters.

Recommendations

  • Run behind a reverse proxy (nginx, Caddy, etc.).
  • Use non-root user inside containers.
  • Enable ASan/TSan/UBSan builds in CI for new changes.
  • Avoid logging sensitive data (queries, user identifiers).
  • Keep dependencies minimal and pinned.
  • Regularly run static analysis (CPPCheck, Clang-tidy).

Reporting vulnerabilities

  • Prefer private, coordinated disclosure.
  • Provide:
    • Affected version
    • Reproduction steps
    • Impact assessment

There aren’t any published security advisories