Threat Model Forge for VS Code

View threat-model diagrams, review security findings, and edit model files without leaving VS Code.

Get Started

1. Install Threat Model Forge by hacks4snacks from VS Code's Extensions view (hacks4snacks.tmforge).
2. Open a .tm7 or .tmforge.json file to edit it in Studio.
3. Run Threat Model Forge: New Model to start a JSON model, or use Open File to open an existing model.

For offline installation, download the VSIX from the GitHub release and run Extensions: Install from VSIX... in the Command Palette. Reload the window after updating.

Use Threat Model Forge: Open Studio or Reopen Editor With to switch a model from the text editor to Studio.

Create with Copilot

With GitHub Copilot Chat enabled, use the normal Agent mode and invoke the bundled tmforge-vscode skill:

/tmforge-vscode Create a draft threat model for this service. Use the current implementation as evidence and keep unknown controls explicit.

Copilot can also load the skill automatically when a request matches its description. There is no separate Tmforge agent to select.

The agent queries the bundled schema and catalogs, creates a new unsaved JSON model, opens it in Studio, and reads it back for analysis. New drafts appear in an Untitled tab without an associated filesystem path. Review the diagram and findings, then use Save to choose a location and a filename ending in .tmforge.json.

For updates, open a model and ask the agent to inspect or change it. Updates use the current document, including unsaved changes, and are undoable. A revision check refuses edits based on stale content. Native TM7 changes use the same source-preserving engine path as Studio. The tools do not explicitly save files; VS Code's Auto Save setting still applies. Deleting objects can remove their dependent threats and decisions, just as in Studio.

The four tools are available to Copilot's general agent: #tmforgeCatalog, #tmforgeCreateModel, #tmforgeInspectModel, and #tmforgeUpdateModel. Inspection and updates only access already-open model documents. Chat results are limited to 1 MiB; larger models remain available in Studio.

The skill instructs Copilot to use tmforge tools for model changes, but does not restrict the current agent's available tools. Validation and revision checks apply to tmforge tool calls; generic file edits can bypass them.

Set tmforge.copilot.enabled to false to disable these contributions. Studio and local analysis continue to work without Copilot. If the skill is missing, check that Copilot Chat and agent mode are available under your organization's policy, the workspace is trusted, and the integration is enabled.

Review Findings

VS Code's Problems panel updates when you open or save a model. Choose Analyze in Studio to review threats, select affected elements, and record triage on the current model.

After editing, save the file to refresh findings. To check unsaved changes, run Threat Model Forge: Analyze Model from the Command Palette or choose Analyze in Studio.

Findings are attached to the model file rather than specific source lines. Rule and element IDs help you identify the affected parts of the model.

Edit Model Files

Studio supports stencil-based drawing, connections, properties, multiple pages, layout, analysis, threat triage, comparison, and reports. See the Studio guide for the editing controls.

Use VS Code's normal Save, Save As, Undo, and Redo commands. Canvas edits mark the document as unsaved; opening or analyzing it does not change the file. Open source switches to text editing. Source edits and changes in another editor update the canvas. Invalid source pauses visual editing until the XML or JSON is corrected.

Open File opens .tm7 and JSON models directly. Importing another format creates a separate unsaved .tmforge.json model after conversion review, leaving the original untouched. Export and Report let you choose a destination for a separate file. URL sharing is not supported in the extension.

Native TM7 Files

Editing a .tm7 keeps it in its original format. Saves preserve unrelated native content, including the embedded template and existing threat data. New edits extend the template when needed. Deleting an element, flow, or page also removes its dependent threats and decisions; Undo restores them.

Opening a native file or saving without changes leaves its bytes unchanged. Edited saves may change XML formatting. Exporting a native .tm7 copy retains the current document, including unsaved edits; exporting to JSON or another format shows any conversion warnings.

Native line trust boundaries are retained but are not drawn on the Studio canvas. Models containing them show a warning because canvas analysis may omit their crossings. Deleting a page removes its hidden objects as well.

JSON Editing

JSON completion and structural validation are available for:

• Model manifests: *.manifest.json and *.tm.json.
• Rule packs: *.tmrules.json.
• Suppression files: *.suppressions.json.

Version 2 rule packs have full schema validation; older rule-only files have basic structural hints. These editing hints do not run custom rules or apply suppressions to a model.

Supported Workflows

Use desktop VS Code 1.138 or later in a trusted workspace, including remote development workspaces. Browser-only VS Code and virtual workspaces are not supported.

Studio analysis uses tmforge's built-in rules and respects the model's disabled-rule selections. In Studio, Analysis Rules can load a custom JSON rule pack for the current document session. Reload that pack after restarting VS Code; packs are not automatically read from the workspace. A model that expects an unavailable pack reports an error. Suppression files are not applied by the extension; use the tmforge CLI for suppression workflows.

Models can contain up to 32 pages, 1,024 elements, and 2,048 diagram lines, with a maximum file size of 8 MiB. More complex models may exceed processing limits. Analysis requests time out after 30 seconds; choose Analyze to retry.

Privacy

Analysis runs on your computer or, when using remote development, on your remote host. The extension does not send model contents to a separate analysis service.

When using Copilot, repository evidence, model content returned by tools, and findings become part of the Copilot conversation and follow its data-handling policies. Local engine execution does not make a Copilot conversation offline or prevent that context from reaching your selected model provider. Use only material you are authorized to share. The extension does not make model-provider requests itself, install software, or change Copilot's approval settings.

Support

Report problems or request improvements in GitHub Issues. Include the extension version, VS Code version, operating system, and the error message. Use a minimal, sanitized model when a reproduction is needed; do not attach confidential architecture or credentials.