Skip to content

Security: IvanMurzak/Context-Engine

SECURITY.md

Security Policy

Context Game Engine follows a coordinated vulnerability disclosure (CVD) process. This page is the reporter-facing policy: how to reach us privately, what to expect, and the safe harbor we extend to good-faith research. The maintainer-side workflow and the EU Cyber Resilience Act (CRA) reporting-readiness note live in docs/cvd-and-cra-readiness.md; the engine's trust-boundary map is docs/security-boundaries-v1.md.

Supported versions

The engine is pre-alpha and has no versioned releases yetmain is the only supported line, and security fixes land there.

Version Supported
main

Once versioned releases exist, this table will enumerate the supported release lines.

Reporting a vulnerability

Do not open a public issue, discussion, or pull request for a suspected vulnerability.

  • Primary (private) channel — GitHub private vulnerability reporting: https://github.com/IvanMurzak/Context-Engine/security/advisories/new. If the report form is unavailable (the repository setting may still be in the process of being enabled), use the fallback below.
  • Fallback: email support@ai-game.dev with the subject prefix [SECURITY]. Please do not include full exploit details in the initial email; we will move the exchange into a private GitHub security advisory.

A useful report includes: the affected component or path (e.g. src/editor/bridge/, src/runtime/js/), the commit or version you tested, reproduction steps or a minimal proof of concept, and your assessment of the impact. Reports about the engine's designed trust posture (untrusted same-user code — see docs/security-boundaries-v1.md) are welcome too; please say which documented boundary you believe is violated.

Response targets (SLOs)

This is a solo-maintainer, pre-alpha project; these are the targets we commit to and expect to meet:

Stage Target
Acknowledgement of your report within 3 business days
Triage + initial severity assessment within 7 calendar days of acknowledgement
Fix, or a documented mitigation plan, for a confirmed vulnerability within 90 days of triage (faster when the vulnerability is being actively exploited)
Coordinated public disclosure default 90 days from the report, or earlier/later by mutual agreement (e.g. when a fix needs more time)

When a report is declined (not a vulnerability, out of scope, works as designed), we say so explicitly with the reasoning, and you are free to disclose after the default window.

A fixed vulnerability is disclosed via a published GitHub security advisory that credits you by name by default — say so in your report (or any time before publication) if you prefer to remain anonymous.

Scope

  • In scope: everything in this repository — EditorKernel, RuntimeKernel, the context CLI, schemas, build tooling (tools/, bench/), and the documented trust boundaries (docs/security-boundaries-v1.md).
  • Out of scope: the separate MCP plugin repositories (Unity-MCP, Godot-MCP, Unreal-MCP — their own repos/policies), the ai-game.dev online services (a separate product), and vulnerabilities purely in third-party dependencies (report upstream — but do tell us as well so we can pin or bump the dependency).

Safe harbor

We support good-faith security research on Context Engine:

  • We will not initiate legal action (including under the Context Engine EULA, LICENSE.md) or report you to law enforcement for security research that is conducted in good faith, complies with this policy, and stays within the scope above.
  • Good faith means: test against your own installations/builds; do not access, destroy, or exfiltrate data that is not yours (a minimal proof of concept is fine); do not degrade infrastructure you do not own; comply with applicable law; and give us the coordinated-disclosure window above before publishing.
  • This safe harbor authorizes security research and disclosure only — it does not extend the EULA's grant otherwise (e.g. it is not a license to redistribute the engine), and it cannot authorize testing against third-party systems or services.

If you are unsure whether something is covered, ask first via the private channel — we would rather answer a question than lose a report.

There aren't any published security advisories