Threat Model Reviewer
Public repository documentation · View source on GitHub
On this page
Find the gaps before your threat model review.
Use the app, CLI and skill from the same product release. Updating a guide does not add new commands to an older installed binary.
Review Microsoft Threat Modeling Tool .tm7 and OWASP Threat Dragon .json
models, work through specific findings, and save a corrected model for human
review. The Windows app, CLI and Copilot CLI skill use the same deterministic
review engine. You do not need Copilot to run a review.
Product site Downloads User guide CLI Skill
Start with a safe example
Download the synthetic customer-portal model and open it with Open .tm7. Under the default rubric it returns NOT READY, 65/100, and 27 gating findings for threats still marked Not Started. The quick-start guide explains the next steps.
For automation, extract the CLI bundle and run:
.\ThreatModelReviewer.Cli.exe .\customer-portal.tm7 --md .\review.md
$LASTEXITCODE
An exit code of 2 is expected for this sample: the review ran successfully and
found gating issues. Code 1 means the command failed, not that a model failed review.
What you can do
| Task | Where | Result |
|---|---|---|
| Review an existing model | App or CLI | 72 deterministic checks, a readiness verdict, score and traceable findings |
| Correct the model | Fix tab or CLI fix |
Preview changes, then write a corrected .tm7; suggestions are not automatically verified mitigations |
| Draft a model | Create/Assistant tabs or CLI authoring commands | Components, flows and a STRIDE starting point to review and refine |
| Start from existing architecture | Create import or CLI ingest |
Extract structure from IaC, supported diagrams and documents |
| Start from deployed Azure | Create > Build from Azure or CLI azure |
Read resource metadata and infer possible access paths; Azure CLI sign-in and read access are required |
| Track progress | Compare, Ask and History | Compare model revisions, ask factual questions and keep local review history |
| Hand off security work | Reports and work-item exports | Share findings and mitigation proposals with the team implementing the controls |
| Use an AI agent | Copilot CLI skill plus the CLI bundle | Ask the agent to run the engine and report its actual output |
Generated threats start as Needs Investigation. A role assignment indicates permission, not observed traffic. Azure discovery is not a complete map of an application: shared-key access, other resource groups and unknown resource types can leave gaps. Inspect the draft and its evidence before relying on it.
Newly generated files need the supported baseline/template workflow when they must reopen in Microsoft TMT. Follow the Create guidance in the user guide; parsing a file in this reviewer alone does not prove TMT compatibility. The compatibility guide distinguishes supported review and authoring workflows from native TMT features and template-dependent behavior.
What the result does not mean
The score describes the model, not a measured reduction in system risk. READY WITH NOTES means there are no gating findings under the selected rubric and policy. It is not Microsoft approval or proof that security controls work.
This application does not patch source code or change deployed infrastructure. It helps identify missing threats, improve the model and hand off actionable work. The owning team implements mitigations and supplies verification evidence.
Choose a download
| Package | Choose it for |
|---|---|
...-x64.msi - recommended |
Desktop installation or managed deployment, per-machine or per-user |
...-win-x64-portable.zip |
Extract-and-run desktop use |
...-setup.exe |
Per-user desktop installer |
...-cli-win-x64.zip |
Scripts and Windows CI runners |
...-skill.zip |
Copilot CLI skill; install the CLI bundle alongside it |
...-x64.msix and ...-publisher.cer |
Experimental MSIX distribution |
Windows 10/11 x64. App and CLI bundles include .NET; the skill is instructions and a CLI resolver, not a replacement runtime. Executables and installers are signed with the project's current self-signed certificate. ZIP files are not Authenticode-signed containers. Check the installation guide and security policy before installing.
Privacy and optional AI
Deterministic model review works locally without network access. Optional AI features send context to the configured provider; image extraction sends the image you select. Azure discovery and enabled MCP data sources use network access separately. Review Data handling before enabling them.
The app does not collect product analytics. Update checks can be disabled. External providers, CLI runtimes and MCP servers have their own terms and settings.
Developer context and evidence
The application includes restricted MCP developer context, a shell-free Azure discovery runner, and captured-revision comparison. Generation retains edited metadata and protects evidence ownership; SDL exports identify the evaluated source bytes and publish the manifest last. Use matching binaries and skill instructions; older releases do not implement every current command or safeguard.
Microsoft Learn, Azure metadata and GitHub repository context are separate, default-off sources. Read-only access is not a privacy guarantee, and none changes the deterministic verdict. Azure DevOps MCP remains deferred, not a shipped option.
Guides and support
Install · User guide · FAQ · CLI · Skill · Architecture · Microsoft TMT compatibility · Enterprise deployment · Third-party notices · Changelog
Report a bug with a check ID and sanitized example. Never attach a real threat model or Azure inventory to a public issue. Report vulnerabilities privately.