Security Policy

Threat Model Reviewer is a defensive security tool that handles highly sensitive input: your threat models. This policy covers how to report a vulnerability, how the product handles data, how it is built and signed, and which versions receive fixes.

• Data handling in full detail: Data handling & privacy
• Managed/offline deployment: Enterprise deployment

Reporting a vulnerability

Please do not open a public issue for security vulnerabilities.

Report security issues privately via GitHub's "Report a vulnerability" button under the Security tab of the public repository (ArasaniRohithReddy/app-releases), or by emailing the maintainer. Include:

• the affected version and install method (portable / installer / MSIX),
• a description of the issue and its impact,
• reproduction steps or a proof of concept.

You can expect an acknowledgement within a few days. Please give a reasonable disclosure window before any public discussion.

What to expect

Researchers who report in good faith will be credited in the advisory unless they prefer to remain anonymous. We will not pursue action against good-faith research that respects user privacy and avoids service disruption.

How the application handles your data

Threat models are sensitive, so the application is designed to keep them local:

• The deterministic review runs entirely on your machine. The verdict, score, and findings never depend on a network call.
• AI features are optional and additive. Finding tools send contextual text; extraction and refinement can also send supplied documents, drafts or images. Text redaction is a safeguard, not a guarantee that an input is free of secrets.
• Prompts go to your own GitHub Copilot seat and are metered against your quota.
• An OpenAI-compatible provider (Azure OpenAI / OpenAI / local) can be used instead of Copilot; the API key is stored locally and DPAPI-protected.
• No product analytics are sent by the app. Updates, account checks, Azure discovery and enabled MCP sources have separate network paths. External runtimes and servers have their own terms and settings; disable integrations for an offline-only workflow.

Every network destination, storage path and redaction rule is enumerated in Data handling & privacy.

Security of the product itself

Because this tool parses untrusted files and passes their content to a language model, it is built defensively:

The deterministic core has no AI dependency, so no model output can alter a verdict, a score, or the contents written back to a .tm7 without explicit user approval of each change.

Integration safeguards

Version 2.6.0 includes restricted/default-off MCP profiles, separate DPAPI-protected GitHub-context credentials, explicit SDK ambient-context opt-outs, bounded connection/process cleanup, and a shell-free direct Azure reader with a captured subscription. These are separate controls: the direct Azure allowlist does not constrain an MCP server.

Generation ownership checks preserve unowned or edited sidecars. Compare and SDL bind provenance to captured source bytes; ambiguous comparison matches are disclosed, and SDL invalidates an older success manifest before replacement. Neither file hashes nor model dispositions verify deployed controls. Exported provenance may contain absolute paths and sensitive metadata; inspect it before sharing.

See MCP, Azure discovery, Compare, and data handling. These safeguards do not sandbox a process or establish full authenticated-service, native UI or installation acceptance. Older installed binaries do not acquire them until updated to a matching release.

Supported versions

Fixes are delivered in the next release rather than as patches to older versions. Because in-app updating has shipped since v1.0.4, staying current is normally automatic.

Code signing

Application executables and installers (MSI, setup .exe, and MSIX) are Authenticode-signed. ZIP containers, Markdown skill files and the public certificate are not Authenticode-signed application binaries. For now they're signed with a self-signed certificate (CN=ArasaniRohithReddy), which means:

• the signature binds the file to the signing key, but does not establish public publisher trust. Verify the expected publisher and integrity metadata;
• MSIX certificate trust requires approval under your organization's policy, as described in the install guide;
• Microsoft Defender SmartScreen or application-control policy may still warn or block execution. Do not dismiss warnings automatically or bypass an administrator block; obtain IT approval when required.

CA-issued, EV or Azure Trusted Signing certificates can establish publisher trust, but do not guarantee the absence of SmartScreen or organizational policy warnings. The release pipeline supports supplying a certificate with -CertPfx or -CertThumbprint to sign application binaries and installers. This migration is planned.