Workspace Trust for Extension Authors

This issue will be updated as information becomes available. Please subscribe and check back.

This issue will function as the landing page for extension authors while Workspace Trust is in development. If you are an extension author, please subscribe to this issue to receive notifications when the issue is updated.

Note that the feature has been in the works for some time now and we are nearing the phase where we would like extension authors to test it out and provide feedback. Once it is ready, we will post the extension authoring guide in this issue so that you can try adopting the new API.

To learn more about the development of the feature, check out our development issue: #106488

Workspace Trust Extension Guide

What is Workspace Trust?

Workspace Trust is a feature driven by the security risks associated with unintended code execution when a user opens a workspace in VS Code. For example, consider that a language extension, in order to provide functionality may execute code from the currently loaded workspace. In this scenario, the user should trust that the contents of the workspace are not malicious. Workspace Trust centralizes this decision within VS Code so that extension authors do not have to handle this themselves. VS Code will offer static declaration and API support to onboard extensions quickly without the need to duplicate code across extensions.

Onboarding

Static Declarations

In your extension's package.json, VS Code will support the following new property:

capabilities:
	untrustedWorkspaces:
		{ supported: true } |
		{ supported: false, description: string } |
		{ supported: 'limited', description: string, restrictedConfigurations?: string[] }

For the supported property, the following values are accepted:

true - The extension is fully supported in an untrusted workspace as it does not need workspace trust to perform any functionality. It will be enabled exactly as before.
false - The extension is not supported in an untrusted worskpace as it cannot function without workspace trust. It will remain disabled until workspace trust is granted.
'limited' - Some features of the extension are supported in an untrusted workspace. Trust-sensitive features should be disabled until workspace trust is granted. The extension can use our API to hide or disable these features. Workspace settings can be gated by trust automatically using the restrictedConfigurations property.

For the description property, a description of why trust is needed must be provided to help the user understand what features will be disabled or what they should review before granting or denying workspace trust. If supported is set to true, this property is ignored.

💡The value for the description property should be added to package.nls.json and then referenced in the package.json file for localisation support.

For the restrictedConfigurations property, an array of configuration settings IDs are provided. When provided, the extension will not be given workspace-defined values for the given settings in an untrusted workspace.

How do I decide how my extension should support untrusted workspaces?

To help extension authors understand what is in scope for Workspace Trust, we have created a list of questions to consider.

Does my extension have a `main` entry point?

If the extension does not have a main entry point (ex: themes, language grammars), the extensions does not require workspace trust. Extension authors do not need to take any action for such extensions as they will continue to function independent whether the workspace is trusted or not.

Does my extension rely on files in the opened workspace to provide features?

This can mean things like settings that can be set by the workspace or actual code in the workspace. If the extension never uses any of the contents of the workspace, it probably doesn't require trust. Otherwise, take a look at the other questions.

Does my extension treat any contents of the workspace as code?

The most common example of this is using a project's workspace dependencies, such as the node modules stored in the local workspace. A malicious workspace might check in a compromised version of the module. Thus, this is a security risk for the user and extension. Auxiliary to this, an extension may rely on JavaScript or other configuration files that control the extension or other modules' behavior. However, many other examples could occur, such as executing an opened code file to determine its output a error reporting.

Does my extension use settings that can be defined in the workspace in any code execution?

Your extension might use settings values as flags to a CLI that your extension executes. If these settings are overridden by a malicious workspace, they could be used as an attack vector against your extension. On the other hand, if the settings' values are only used to detect certain conditions, (e.g. if the value of a preferred shell setting is bash or pwsh determines what documentation you show), then it may not be a security risk and does not require Workspace Trust. Below is a section dedicated to guidance on settings to help you find the optimal configuration for your extension.

This is not an exhaustive list of cases that might require Workspace Trust. As we review more extensions, we will update this list. Use this list to think of similar behavior your extension might be doing when considering Workspace Trust.

What if I don't make changes to my extension?

As mentioned above, an extension that does not contribute anything to their package.json will be treated as not supporting untrusted workspaces. It will be disabled when the workspace is not trusted and the user will be notified that some extensions are not working due to Workspace Trust. This measure is the most security-conscious approach for the user. Even though this is the default, it is still a best practice to set the appropriate value indicating that as an extension author, you have made the effort to protect the user and your extension from malicious workspace content.

How do I use the API?

As described above, the first step to using the API is adding the static declarations to your package.json. The easiest method of onboarding would be to use a false value for the supported property. Once again, this is the default behavior even if you do nothing, but it's a great signal to the user that you have made a deliberate choice. In this case, your extension does not need to do anything else. It will simply not be activated until trust is given and your extension can know that it is executing with the consent of the user. However, if your extension only requires trust for part of its functionality, this is likely not the best option.

For extensions that wish to gate their features on Workspace Trust, they should use the limited value for the supported property, and we provide the following API:

export namespace workspace {
	/**
		* When true, the user has explicitly trusted the contents of the workspace.
		*/
	export const isTrusted: boolean;

	/**
		* Event that fires when the current workspace has been trusted.
		*/
	export const onDidGrantWorkspaceTrust: Event<void>;
}

We provide a property to determine if the current workspace is trusted and an event for listening to when trust has been granted to the workspace. You can use these block specific code paths and perform any necessary registrations once the workspace has been trusted.

We also expose a context key isWorkspaceTrusted for use in when clauses as described below.

What about my contributions?

Commands, Views, or other UI

When the user has not trusted the workspace, they will operating in a mode with functionality geared towards browsing code. Any features that you provide that will not be trusted should be hidden from the user. This can be done via when clauses and the context key isWorkspaceTrusted. Since code can still call a command even if it is not presented in the UI, you should still block execution or not register a command based on the API above in your extension code.

Configurations/Settings

First, you should review your settings to determine if they need to take trust into account. As described above, a workspace may define a value for a setting that your extension consumes that is malicious to the use. If you identify settings that are vulnerable, you should use 'limited' for the supported property and list the setting ID in the restrictedConfigurations array.

When you add a setting ID to the restrictedConfigurations array, VS Code will only return the user-defined value of the setting in an untrusted workspace. This means your extension needs no additional code changes to handle the setting. When trust is granted, a configuration change event will fire in addition to the workspace trust event.

What is the guidance for debug extensions?

VS Code will prevent debugging in an untrusted workspace. For this reason, generally, debugging extensions do not need to additionally require trust and should select true for the supported property. However, if your extension provides additional functionality, commands, or settings that are not part of the built-in debugging flow, you should use 'limited' and the above guidance.

What is the guidance for task providers?

Similar to debugging, VS Code will prevent running tasks in an untrusted workspace. If your extension provides additional functionality, commands, or settings that are not part of the built-in tasks flow, you should use 'limited' and the above guidance. Otherwise, you can specify true.

How do I test Workspace Trust?

Please see this issue for current details on development of Workspace Trust. It should be up to date with how the feature is currently enabled.

For those subscribed, we are now ready to share the API and guidance for onboarding to workspace trust for extension authors. This is the first time sharing the guidance to a broader audience so please post questions here and we can update it for clarity. Thanks for your engagement.

Is Workspace Trust going to land in the upcoming release?

The setting to enable the feature will remain disabled by default in the upcoming release

SteVen Batten (@sbatten)

SteVen Batten (@sbatten) thank you for providing guidance on how trusted workspaces would work.

I am concerned about your default behaviour and my expectations would be that extensions that provide web views, side bar views, pseudo terminals and commands you can run in terminal, and work with workspace files, would still function as before when this feature is released.

I would rather see this feature as opt-in for extensions that want to be enabled for trusted workspaces, and work as before for untrusted workspaces. I think the behavior you describe is the opposite.

What if I don't make changes to my extension?

As mentioned above, an extension that does not contribute anything to their package.json will be treated as not supporting untrusted workspaces. It will be disabled when the workspace is not trusted and the user will be notified that some extensions are not working due to Workspace Trust. This measure is the most security-conscious approach for the user. Even though this is the default, it is still a best practice to set the appropriate value indicating that as an extension author, you have made the effort to protect the user and your extension from malicious workspace content.

SteVen Batten (@sbatten)

SteVen Batten (@sbatten) Is there a way to prompt the user, "To enable this feature, you must trust the workspace"? For extensions that provide actions a user can execute, simply hiding (and disabling) that functionality may be bad UX. For example, the ESLint extension - at the moment - will prompt the user to execute ./node_modules/.bin/eslint if it is present, saying something like "Do you trust this?" How should that be updated for Workspace Trust? As I read your explanation, the extension should silently disable that functionality for untrusted workspaces. As a user, I would prefer a prompt - "The extension eslint wants to do <some thing> that requires you to trust this workspace - continue?"

As another example, I have an extension that provides code lenses for profiling benchmarks (CPU/memory use). If I hide these code lenses in untrusted workspaces, I'm sure someone will report that as a bug ("the benchmark code lens doesn't show up for my workspace"). I'd like the API to include something like this:

export namespace workspace {
    export function shouldExecuteActionRequiringTrust(description: string, alertOnFailure: bool, details?: string): Promise<bool>
}

If the workspace is not yet trusted and has not yet been prompted, this would show a prompt asking if the user wants to trust the workspace. If the workspace is untrusted and alertOnFailure is true, this would show a notification that, " cannot execute because the workspace is not trusted".

Taras Novak (@RandomFractals)

Taras Novak (@RandomFractals) Ethan Reesor (@firelizzard18) Thanks for your engagement.

Taras Novak (@RandomFractals) you are correct in your understanding of the behavior. It is indeed opt-out, not opt-in. As this is a security feature, we must make the security-conscious decision to disable extensions that execute code and do not tell us about their trust support. We provide a user setting to override whether an extension supports untrusted workspaces locally to unblock users that have extensions that are slow to adopt. This puts the control in the hands of the user and is safe by default.

Ethan Reesor (@firelizzard18) We hear this feedback. It is something we considered when coming up with the current strategy and we will keep our ears to ground on its necessity. With our current UX, we attempt to guide the user into one of two modes of operation, and they must make an explicit choice. In this way, we market the restricted mode for code browsing and navigation, but not active development since active development typically requires execution and thus trust.

We anticipate that a majority of users workflows are with trusted folders and thus should not choose restricted mode to do active development.

For this reason, the guidance is:

If you cannot provide a coherent browsing experience in an untrusted workspace, simply set your support value to false and rely on Code to indicate that features are disabled due to this choice.
If you can provide a solid experience for the restricted mode scenario, do so with limited or true.

SteVen Batten (@sbatten)

SteVen Batten (@sbatten) thanks for your follow up. Sounds like the default behavior has already been decided on. I get why you'd want to configure extension activation and execution this way from security standpoint for untrusted workspaces.

What concerns me most about this approach is that it will create a potential stream of updates and some rework for extension developers similar to how we had to scramble and handle webview resources last year. I can also see a lot of users filing issues in extensions related to this new feature if configuration of this functionality is not well exposed in vscode UI for extensions and documented when you release it.

we must make the security-conscious decision to disable extensions that execute code and do not tell us about their trust support. We provide a user setting to override whether an extension supports untrusted workspaces locally to unblock users that have extensions that are slow to adopt. This puts the control in the hands of the user and is safe by default.

That part sounds like it might be ok, if you also provides proper prompts and notifications when switching between trusted and untrested workspaces.

I am not sure if Eric Amodio (@eamodio) still does his monthly extension developers calls, but I think it would be great if we could do one for this functionality and you show us the whole flow of how it functions, and we hear other extension developers provide feedback that way before you realease this feature. There are only a few devs who expressed interest in learning more about this feature. So, should be manageable to get a few devs on a call for ext. devs review.

Hello, thanks for providing this information and way to follow as we prepare for the roll-out.

Can you paste some screenshots showing the user facing part of it? It would help me understand what choices I make as an extension developer.

I do not think there is a way other than bumping vscode engine version.

SteVen Batten (@sbatten)

SteVen Batten (@sbatten) What is the motivation behind asking debug extensions to mark themselves as safe for untrusted workspaces?
Debugging untrusted code is among the least safe things one could do. Wouldn't it make more sense to mark debug extensions as not supported for untrusted workspaces, just to be double-sure?

SteVen Batten (@sbatten)

SteVen Batten (@sbatten) What is the motivation behind asking debug extensions to mark themselves as safe for untrusted workspaces?
Debugging untrusted code is among the least safe things one could do. Wouldn't it make more sense to mark debug extensions as not supported for untrusted workspaces, just to be double-sure?

vadimcn I think that's covered already. From the description in this comment:

What is the guidance for debug extensions?

VS Code will prevent debugging in an untrusted workspace. For this reason, generally, debugging extensions do not need to additionally require trust and should select true for the supported property. However, if your extension provides additional functionality, commands, or settings that are not part of the built-in debugging flow, you should use 'limited' and the above guidance.

So I don't think debug extensions will be able to debug in an untrusted workspace, no matter what they mark themselves for.

Yes, I've seen that comment. I still don't understand the motivation for asking debug extensions to set "supported": true. Is there any behavioral difference between this and "supported": false?

That makes sense. I think it will have to be up to extension authors to determine based on the guidance above. If the only thing an extension provides is debugging, I think that is functionally equivalent to "supported": false, but there could be other features that are safe in an untrusted workspace (or features that aren't, and "supported": "limited" is appropriate).

feat(extension): Update untrusted workspace support from 'false' to '…

Sunbelt Computer Software

PL/B Language Development and Support

Navigation Menu

Repository navigation

Workspace Trust for Extension Authors #120251

Workspace Trust Extension Guide

What is Workspace Trust?

Onboarding

Static Declarations

How do I decide how my extension should support untrusted workspaces?

Does my extension have a `main` entry point?

Does my extension rely on files in the opened workspace to provide features?

Does my extension treat any contents of the workspace as code?

Does my extension use settings that can be defined in the workspace in any code execution?

What if I don't make changes to my extension?

How do I use the API?

What about my contributions?

Commands, Views, or other UI

Configurations/Settings

What is the guidance for debug extensions?

What is the guidance for task providers?

How do I test Workspace Trust?

What if I don't make changes to my extension?

42 remaining items

What is the guidance for debug extensions?

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Sunbelt Computer Software

PL/B Language Development and Support

Workspace Trust for Extension Authors #120251

Description

Activity

sbatten commented on Apr 23, 2021

Workspace Trust Extension Guide

What is Workspace Trust?

Onboarding

Static Declarations

How do I decide how my extension should support untrusted workspaces?

Does my extension have a main entry point?

Does my extension rely on files in the opened workspace to provide features?

Does my extension treat any contents of the workspace as code?

Does my extension use settings that can be defined in the workspace in any code execution?

What if I don't make changes to my extension?

How do I use the API?

What about my contributions?

Commands, Views, or other UI

Configurations/Settings

What is the guidance for debug extensions?

What is the guidance for task providers?

How do I test Workspace Trust?

sbatten commented on Apr 23, 2021

gjsjohnmurray commented on Apr 23, 2021

sbatten commented on Apr 23, 2021

RandomFractals commented on Apr 24, 2021

What if I don't make changes to my extension?

firelizzard18 commented on Apr 24, 2021

sbatten commented on Apr 27, 2021

RandomFractals commented on Apr 27, 2021

PEZ commented on Apr 28, 2021

42 remaining items

sandy081 commented on Jul 7, 2021

vadimcn commented on Jul 27, 2021

bwateratmsft commented on Jul 28, 2021

What is the guidance for debug extensions?

vadimcn commented on Jul 28, 2021

bwateratmsft commented on Jul 28, 2021

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

Does my extension have a `main` entry point?