If you discover a security vulnerability in Quorate, please report it privately by emailing umutkorkmaz@outlook.com.tr. Include enough detail to reproduce the issue. Please do not open a public issue for security reports.

You can expect an acknowledgement of your report, and we will keep you informed as we investigate and prepare a fix.

Quorate drives AI CLIs that live on your machine, so provider isolation is a core part of its design:

• Opt-in providers. Real CLI providers are disabled by default; only the built-in heuristic runs with zero setup. You enable real reviewers explicitly.
• No shell. Providers are spawned directly, never through a shell, so there is no shell-injection surface.
• Explicit headless args. Each provider runs with explicit headless arguments; empty args are refused so no interactive session is ever opened.
• Dangerous-flag denylist. Session/resume and --dangerously*/--yolo-style flags are rejected unless a profile explicitly opts in with allowDangerousArgs.
• Byte and time caps. Prompts and output are bounded by maxInputBytes / maxOutputBytes, and runtime is bounded by timeoutMs with a forced-kill grace period.
• Scrubbed environment. Providers receive a scrubbed environment built from an explicit allowlist rather than the caller's full environment.

The Action loads .quorate.yml from the pull request's base branch, never from the PR head. A pull request therefore cannot supply the configuration that governs its own review.