---
name: tm7-threat-model
description: 'Creates valid Microsoft Threat Modeling Tool (.tm7) files compatible with the Microsoft Threat Modeling Tool v7.3+. Use this skill whenever asked to create, generate, or modify a .tm7 threat model file, or when performing STRIDE threat modeling that should output a .tm7 file that opens cleanly in the Microsoft Threat Modeling Tool.'
---

# Microsoft Threat Modeling Tool (.tm7) Generator

You generate **valid `.tm7` files** for the Microsoft Threat Modeling Tool (v7.3+). A `.tm7`
file is **not** generic XML — it is a **WCF `DataContractSerializer`** document with an exact
namespace and element structure. If the structure is wrong, the tool refuses to open the file
with:

> "File is not an actual threat model or the threat model may be corrupted."

Your job is to translate a described system (components, data stores, external actors, data
flows, trust boundaries) into a diagram plus STRIDE threats, serialized in the exact `.tm7`
format described below.

## Workflow

When asked to produce a `.tm7` file:

1. **Model the system.** Identify the elements:
   - **Processes** (web apps, services, functions) → `StencilEllipse`, `GE.P`
   - **Data stores** (databases, caches, queues, blobs) → `StencilParallelLines`, `GE.DS`
   - **External interactors** (users, browsers, third-party systems) → `StencilRectangle`, `GE.EI`
   - **Trust boundaries** → `BorderBoundary`, `GE.TB`
   - **Data flows** connecting the above → `Connector`, `GE.DF`
2. **Assign a unique lowercase UUID** (e.g. `148ade68-5c80-40f3-8e1f-4e2cabdb5991`) to every
   stencil and every flow. Never use human-readable ids like `users-browser`.
3. **Lay out coordinates** (`Left`/`Top`/`Width`/`Height`) so stencils don't overlap.
4. **Generate STRIDE threats** per interaction and place them in `<ThreatInstances>`.
5. **Serialize** using the structure in this guide, mirroring `assets/example-minimal.tm7`.
6. **Validate** against the "Common Mistakes" checklist before returning the file.
7. **Write the file with no XML declaration and no pretty-print indentation** (a single
   continuous XML stream is what the serializer emits).

Always open [`assets/example-minimal.tm7`](./assets/example-minimal.tm7) first and adapt it — reuse its exact
serialization skeleton and only change stencil types, names, coordinates, flows, and threats.

## CRITICAL: Serialization format

TM7 files use **WCF `DataContractSerializer` XML**, not standard XML.

The file MUST start with this exact root element — **no `<?xml?>` declaration**:

```xml
<ThreatModel xmlns="http://schemas.datacontract.org/2004/07/ThreatModeling.Model" xmlns:i="http://www.w3.org/2001/XMLSchema-instance">
```

**NEVER use:**
- `<?xml version="1.0" encoding="utf-8"?>` — causes deserialization failure.
- `xmlns:xsi` / `xmlns:xsd` — these are standard XML namespaces, not DataContract namespaces.
- Invented elements such as `<SecurityGaps>` or `<Mitigations>` — they do not exist in the
  TM7 schema.

> **Note:** `<MetaInformation>` (with children like `<Owner>`, `<Contributors>`,
> `<Reviewer>`, `<Assumptions>`, `<ExternalDependencies>`, `<HighLevelSystemDescription>`,
> `<ThreatModelName>`), `<Notes>`, and `<KnowledgeBase>` **are** part of the real schema and
> are emitted by the tool — keep them (see the structure below and `assets/example-minimal.tm7`).
> Just don't invent elements that the tool never produces.

## Required namespace prefixes

| Prefix | URI | Used for |
|--------|-----|----------|
| (default) | `http://schemas.datacontract.org/2004/07/ThreatModeling.Model` | Root `ThreatModel` |
| `xmlns:i` | `http://www.w3.org/2001/XMLSchema-instance` | Type attributes |
| `xmlns:z` | `http://schemas.microsoft.com/2003/10/Serialization/` | Reference ids (`z:Id`) |
| `xmlns:a` | `http://schemas.microsoft.com/2003/10/Serialization/Arrays` | Arrays / collections |
| `xmlns:b` | `http://schemas.datacontract.org/2004/07/ThreatModeling.KnowledgeBase` | Stencil properties |
| `xmlns:c` | `http://www.w3.org/2001/XMLSchema` | Primitive type values |

## File structure (correct order)

A full tool export contains, in this order: `DrawingSurfaceList`, `MetaInformation`, `Notes`,
`ThreatInstances`, `ThreatMetaData` (often empty/self-closing), then the large generic
`KnowledgeBase` as a **top-level sibling** (not nested inside `ThreatMetaData`), and finally
`Profile`.

```xml
<ThreatModel xmlns="..." xmlns:i="...">
  <DrawingSurfaceList>
    <DrawingSurfaceModel z:Id="i1" xmlns:z="...">
      <GenericTypeId xmlns="...Abstracts">DRAWINGSURFACE</GenericTypeId>
      <Guid xmlns="...Abstracts">{guid}</Guid>
      <Properties xmlns="...Abstracts" xmlns:a="...Arrays">...</Properties>
      <TypeId xmlns="...Abstracts">DRAWINGSURFACE</TypeId>
      <Borders xmlns:a="...Arrays">
        <!-- Stencil elements: processes, data stores, external entities, boundaries -->
      </Borders>
      <Lines xmlns:a="...Arrays">
        <!-- Data flow lines connecting stencils -->
      </Lines>
      <Notes xmlns:a="...Arrays"/>
    </DrawingSurfaceModel>
  </DrawingSurfaceList>
  <MetaInformation>
    <!-- Owner, Contributors, Reviewer, Assumptions, ThreatModelName, etc. -->
  </MetaInformation>
  <Notes xmlns:a="...Arrays"/>
  <ThreatInstances>
    <!-- Threat entries -->
  </ThreatInstances>
  <ThreatMetaData/>
  <KnowledgeBase z:Id="i21" xmlns:a="...ThreatModeling.KnowledgeBase" xmlns:z="...">
    <!-- Generic SDL stencil/threat catalog — top-level sibling of ThreatMetaData -->
  </KnowledgeBase>
  <Profile>
    <PromptedKb xmlns=""/>
  </Profile>
</ThreatModel>
```

> The `<KnowledgeBase>` (the generic SDL stencil/threat catalog) is large but **required** —
> the tool uses it to resolve every stencil `TypeId`. It is a **top-level sibling** placed after
> `ThreatMetaData` and before `Profile`, **not** nested inside `ThreatMetaData`. Reuse it verbatim
> from `assets/example-minimal.tm7`; only add stencils whose `TypeId` already appears in that
> KnowledgeBase.

## Stencil elements

Each stencil in `<Borders>` is wrapped in `<a:KeyValueOfguidanyType>`:

```xml
<a:KeyValueOfguidanyType>
  <a:Key>{guid}</a:Key>
  <a:Value z:Id="i2" i:type="StencilEllipse">
    <GenericTypeId xmlns="...Abstracts">GE.P</GenericTypeId>
    <Guid xmlns="...Abstracts">{guid}</Guid>
    <Properties xmlns="...Abstracts">
      <a:anyType i:type="b:HeaderDisplayAttribute" xmlns:b="...KnowledgeBase">
        <b:DisplayName>Web Application</b:DisplayName>
        <b:Name/>
        <b:Value i:nil="true"/>
      </a:anyType>
      <a:anyType i:type="b:StringDisplayAttribute" xmlns:b="...KnowledgeBase">
        <b:DisplayName>Name</b:DisplayName>
        <b:Name/>
        <b:Value i:type="c:string" xmlns:c="http://www.w3.org/2001/XMLSchema">My Component</b:Value>
      </a:anyType>
      <!-- Out Of Scope, Reason, configurable attributes -->
    </Properties>
    <TypeId xmlns="...Abstracts">SE.P.TMCore.WebApp</TypeId>
    <Height xmlns="...Abstracts">100</Height>
    <Left xmlns="...Abstracts">400</Left>
    <StrokeDashArray i:nil="true" xmlns="...Abstracts"/>
    <StrokeThickness xmlns="...Abstracts">1</StrokeThickness>
    <Top xmlns="...Abstracts">200</Top>
    <Width xmlns="...Abstracts">100</Width>
  </a:Value>
</a:KeyValueOfguidanyType>
```

### Stencil shape types

| Shape | `i:type` | `GenericTypeId` | Description |
|-------|----------|-----------------|-------------|
| Process (circle) | `StencilEllipse` | `GE.P` | Processes, web apps, services |
| Data store (parallel lines) | `StencilParallelLines` | `GE.DS` | Databases, storage, caches |
| External interactor (rectangle) | `StencilRectangle` | `GE.EI` | Users, external systems |
| Trust boundary | `BorderBoundary` | `GE.TB` | Trust boundaries |

### Common `TypeId` values (SDL TM knowledge base)

| `TypeId` | Component |
|----------|-----------|
| `SE.P.TMCore.WebApp` | Web Application |
| `SE.P.TMCore.AzureAppServiceWebApp` | Azure App Service Web App |
| `SE.P.TMCore.AzureEventHub` | Azure Event Hub |
| `SE.P.TMCore.DynamicsCRM` | Dynamics CRM |
| `SE.DS.TMCore.SQL` | SQL Database |
| `SE.DS.TMCore.AzureSQLDB` | Azure SQL Database |
| `SE.EI.TMCore.Browser` | Browser |
| `SE.EI.TMCore.Mobile` | Mobile Client |

## Data flow lines

Lines in `<Lines>` also use `<a:KeyValueOfguidanyType>`, with `i:type="Connector"`:

```xml
<a:KeyValueOfguidanyType>
  <a:Key>{line-guid}</a:Key>
  <a:Value z:Id="i10" i:type="Connector">
    <GenericTypeId xmlns="...Abstracts">GE.DF</GenericTypeId>
    <Guid xmlns="...Abstracts">{line-guid}</Guid>
    <Properties xmlns="...Abstracts">...</Properties>
    <TypeId xmlns="...Abstracts">SE.DF.TMCore.Request</TypeId>
    <HandleX xmlns="...Abstracts">0</HandleX>
    <HandleY xmlns="...Abstracts">0</HandleY>
    <SourceGuid xmlns="...Abstracts">{source-stencil-guid}</SourceGuid>
    <SourceX xmlns="...Abstracts">0</SourceX>
    <SourceY xmlns="...Abstracts">0</SourceY>
    <TargetGuid xmlns="...Abstracts">{target-stencil-guid}</TargetGuid>
    <TargetX xmlns="...Abstracts">0</TargetX>
    <TargetY xmlns="...Abstracts">0</TargetY>
  </a:Value>
</a:KeyValueOfguidanyType>
```

## Property attribute types

Properties use typed `<a:anyType>` elements:

| `i:type` | Purpose | Value |
|----------|---------|-------|
| `b:HeaderDisplayAttribute` | Section header | `i:nil="true"` |
| `b:StringDisplayAttribute` | Text value (Name, Reason) | `i:type="c:string"` |
| `b:BooleanDisplayAttribute` | Boolean (Out Of Scope) | `i:type="c:boolean"` |
| `b:ListDisplayAttribute` | Dropdown list | Has `<b:SelectedIndex>` |

## Threat instances

Threats go in `<ThreatInstances>` using `<a:KeyValueOfstringThreatpc_P0_PhOB>` (note the exact
`PhOB` suffix). Unlike stencils, the threat `<a:Value>` fields are **`b:`-prefixed** (the
`ThreatModeling.KnowledgeBase` namespace), and the `<a:Key>` is the literal concatenation
`TH<id> + <SourceGuid> + <FlowGuid> + <TargetGuid>`:

```xml
<ThreatInstances xmlns:a="...Arrays">
  <a:KeyValueOfstringThreatpc_P0_PhOB>
    <a:Key>TH117{source-guid}{flow-guid}{target-guid}</a:Key>
    <a:Value xmlns:b="...KnowledgeBase">
      <b:ChangedBy/>
      <b:DrawingSurfaceGuid>{drawing-surface-guid}</b:DrawingSurfaceGuid>
      <b:FlowGuid>{flow-guid}</b:FlowGuid>
      <b:Id>32</b:Id>
      <b:InteractionKey>{source-guid}:{flow-guid}:{target-guid}</b:InteractionKey>
      <b:InteractionString i:nil="true"/>
      <b:ModifiedAt>2025-01-01T00:00:00</b:ModifiedAt>
      <b:Priority>High</b:Priority>
      <b:Properties>
        <a:KeyValueOfstringstring>
          <a:Key>Title</a:Key>
          <a:Value>An adversary may spoof the user and gain access</a:Value>
        </a:KeyValueOfstringstring>
        <a:KeyValueOfstringstring>
          <a:Key>UserThreatCategory</a:Key>
          <a:Value>Spoofing</a:Value>
        </a:KeyValueOfstringstring>
        <a:KeyValueOfstringstring>
          <a:Key>UserThreatShortDescription</a:Key>
          <a:Value>Spoofing is when a process or entity is something other than its claimed identity.</a:Value>
        </a:KeyValueOfstringstring>
        <a:KeyValueOfstringstring>
          <a:Key>PossibleMitigations</a:Key>
          <a:Value>Enable multi-factor authentication and least-privilege access control.</a:Value>
        </a:KeyValueOfstringstring>
        <a:KeyValueOfstringstring>
          <a:Key>Priority</a:Key>
          <a:Value>High</a:Value>
        </a:KeyValueOfstringstring>
        <a:KeyValueOfstringstring>
          <a:Key>SDLPhase</a:Key>
          <a:Value>Design</a:Value>
        </a:KeyValueOfstringstring>
      </b:Properties>
      <b:SourceGuid>{source-stencil-guid}</b:SourceGuid>
      <b:State>Mitigated</b:State>
      <b:StateInformation i:nil="true"/>
      <b:TargetGuid>{target-stencil-guid}</b:TargetGuid>
      <b:Title i:nil="true"/>
      <b:TypeId>TH117</b:TypeId>
      <b:Upgraded>false</b:Upgraded>
      <b:Wide>false</b:Wide>
    </a:Value>
  </a:KeyValueOfstringThreatpc_P0_PhOB>
</ThreatInstances>
```

**Every GUID must resolve:** `SourceGuid` and `TargetGuid` must equal `<a:Key>` values of real
stencils in `<Borders>`, and `FlowGuid` must equal the `<a:Key>` of a real connector in
`<Lines>`. Dangling references produce a model that opens with missing diagram elements.

Use the standard STRIDE categories for `UserThreatCategory`: **S**poofing, **T**ampering,
**R**epudiation, **I**nformation Disclosure, **D**enial of Service, **E**levation of Privilege.

## Common mistakes that break TM7 files

1. **Adding an `<?xml version="1.0"?>` declaration** — `DataContractSerializer` does not emit one.
2. **Using `xmlns:xsi` / `xmlns:xsd`** instead of DataContract namespaces.
3. **Using simple element names** like `<Border>`, `<Line>`, `<Stencil>` — you must use the
   DataContract wrapper types such as `<a:KeyValueOfguidanyType>`.
4. **Inventing elements the tool never emits** like `<SecurityGaps>` or `<Mitigations>` — these
   are not in the schema. (`<MetaInformation>`, `<Notes>`, and `<KnowledgeBase>` **are** valid
   and must be preserved.)
5. **Using human-readable GUIDs** like `users-browser` instead of real UUIDs
   (e.g. `148ade68-5c80-40f3-8e1f-4e2cabdb5991`).
6. **Dangling references** — a `Line`, threat `SourceGuid`/`TargetGuid`, or threat `FlowGuid`
   that points to a stencil/flow GUID that isn't actually defined in `<Borders>`/`<Lines>`.
   Every reference must resolve to an included element.
7. **Missing or duplicated `z:Id` reference attributes** — every serialized object needs a
   `z:Id`, and each `z:Id` (e.g. `i1`, `i2`, `i10`) must be **unique** across the whole file.
   When you duplicate a template block to add an element, always renumber its `z:Id` (and any
   nested ones) to values not used elsewhere; reusing an id creates duplicate DataContract
   object ids and makes deserialization fail.
8. **Missing the `xmlns` on child elements** — each `GenericTypeId`, `Guid`, `Properties`,
   `TypeId`, etc. must carry its own
   `xmlns="http://schemas.datacontract.org/2004/07/ThreatModeling.Model.Abstracts"`.
9. **Pretty-printing with indentation** — the correct output is a single continuous XML stream
   with no added newlines or indentation inside the content.

## Reference asset

Always use [`assets/example-minimal.tm7`](./assets/example-minimal.tm7) in this skill's
directory as the structural reference. It is a fully synthetic, sanitized export (no personal or
project data) that opens cleanly in the tool: two stencils connected by one data flow, with one
STRIDE threat whose every reference resolves. Adapt the stencil types, names, properties,
coordinates, data flows, and threats to the user's architecture, but **never** change the
serialization format or namespace structure, and only use stencil `TypeId` values that already
appear in its bundled `KnowledgeBase`. After generating, mentally diff your output's skeleton
against the example to confirm every namespace, wrapper element, and GUID reference matches.