Microsoft Scout is a desktop AI application currently available through the Frontier preview program. It can execute shell commands directly on your machine, including git, gh, curl, and PowerShell, based on plain text requests instead of typed syntax. This guide covers how to run these commands and how to control what Scout is allowed to do without asking first.
What You Need Before Starting
Scout does not open to every Microsoft 365 user yet. The requirements below apply before you can sign in and start running shell commands.
| Requirement | Detail |
|---|---|
| Platform | Windows 11 or macOS 12 (Monterey) or later |
| Account | Microsoft 365 work or school account (personal accounts unsupported) |
| Microsoft 365 Copilot | Active license assigned to your account |
| GitHub Copilot | Business or Enterprise license |
| Windows library | Latest Microsoft Visual C++ redistributable |
| Install rights | Local administrator permissions and an Intune-enabled account |
Your IT admin needs to complete the Scout access setup and enroll in the Frontier program before you can sign in. Once that is done, download and run the Scout installer for your platform, then sign in with your organizational Microsoft 365 credentials and your GitHub Copilot license. Since Scout is still in development, some capabilities and default behaviors may change in future releases.
Scout also works within a single workspace directory you assign it, so its shell activity stays contained to that folder and cannot touch files or projects outside it.
Every command Scout runs falls into one of three permission tiers: auto-approve, prompt, or deny. This controls whether a command runs instantly or waits for you to confirm it. The methods below point out when this comes into play, and Method 3 covers the full breakdown.
Method 1: Ask Scout to Run a Git Command
Scout accepts git requests as natural language instructions typed into its chat interface.
Open Microsoft Scout and select the chat input at the bottom of the window. Type a request describing what you want done, then press Enter or select Send. Scout streams its response in real time, including any command output.
Examples that work directly:
- “Show me the diff for the last 3 commits”
- “Create a branch and commit my changes”
- “List my open PRs”
- “Create a PR with this description”
Read-only git commands such as git log and git diff run automatically without a prompt. Commands that change repository state, such as git push, pause and show you the exact command before it runs. Method 3 explains why this happens.
Method 2: Run Builds, Tests, and Linters
Scout can execute common development tasks the same way it handles git requests.
Type a request describing the task instead of running the command yourself. Scout builds the correct CLI syntax and executes it in your workspace directory.
Examples:
- “Build the project and show me any errors”
- “Run the test suite and report failures”
- “Lint the src directory and fix any issues”
- “Run the failing test with verbose output and explain the error”
Method 3: Understand the Permission Tiers
Every shell command Scout runs falls into one of three tiers, and this determines whether it runs immediately or waits for your approval.
| Tier | Behavior | Examples |
|---|---|---|
| Auto-approve | Runs without prompting | ls, cat, grep, git log, git diff, docker ps, npm list, gh pr list, kubectl get |
| Prompt | Pauses for your approval | npm install, git push, curl, network requests, file writes |
| Deny | Blocked entirely | Destructive commands such as rm -rf /, format |
When a command needs approval, Scout displays the exact command before running it. Select Approve to proceed or Deny to block it. You can also select Always allow to auto-approve similar commands going forward. This saves time on repetitive work, but it also turns future matching commands into silent actions, so use it only for commands you are confident are safe to repeat.
Method 4: Customize Shell Permissions
The Always allow option in Method 3 is really a shortcut for the same settings you can configure directly. If you would rather set these rules up front instead of approving commands one at a time, you can change which commands auto-approve, which require a prompt, and which are blocked outright.
Go to Settings > Permissions to open the permissions editor. From here you can add patterns to the allow list so specific commands run without prompting, such as npm test or python *.py. The * acts as a wildcard, so python *.py matches any command that runs a Python file regardless of the filename. You can also add patterns to the deny list to block entire categories of commands using the same wildcard format.
Toggle Auto-approve read-only on or off depending on how much oversight you want over read-only operations.
Method 5: Mark Sensitive Paths
Allow and deny patterns control which commands run, but some folders need protection regardless of which command touches them. Certain directories or files may need approval every time, even when the underlying command would normally auto-run.
In Settings > Permissions, add directory or file paths that require explicit approval before Scout can read or write to them. This applies even if the operation would otherwise be classified as auto-approve.
Method 6: Turn Off Shell Access Entirely
Sensitive paths and permission patterns give you fine-grained control, but if you want to remove shell access altogether instead of managing individual rules, you can disable the capability at the category level.
Go to Settings > Permissions and toggle the Shell category off. Once disabled, Scout cannot use shell tools at all, and they no longer appear in its system prompt.
Method 7: Hand Off Heavier Work to Sub-Agents
For larger or parallel jobs, Scout can launch specialized sub-agents that run in the background and report back once they finish, instead of tying up your main conversation.
| Agent type | What it does | When to use |
|---|---|---|
| Explore | Fast codebase research across many files and modules | Understanding unfamiliar code, tracing data flows, analyzing architecture |
| Task | Runs builds, tests, lints, and installations | Running a test suite, building a project, installing dependencies |
| Code review | Analyzes code changes for bugs and security issues | Reviewing diffs, checking for logic errors before committing |
| Research | Thorough web and repository searches with citations | Investigating technical questions, comparing libraries, finding docs |
| General-purpose | Full toolset for complex multi-step work | Refactoring across files, implementing features, complex migrations |
Scout decides on its own when a job would benefit from parallel execution or an isolated context, then launches one or more sub-agents with specific instructions. This lets something like a full test run or a diff review happen in the background without blocking whatever you ask it next. Since these jobs finish out of view, knowing how to confirm the result matters more here than with a command you watch run in real time.
How to Confirm a Command Actually Ran
Scout streams its progress into the conversation as it works, including the tool calls it makes, any permission prompts, files it creates, and the final results. Output renders in rich markdown, so build errors, test failures, and command output come back formatted with headings, tables, and syntax-highlighted code blocks.
For commands that write to disk, files Scout creates or edits are saved directly to your workspace directory. If a git operation such as a branch creation or commit was in the Prompt tier, seeing your Approve confirmation followed by the command’s output in the conversation confirms it went through.
A few questions come up often once you start using Scout day to day.
Frequently Asked Questions
How is Microsoft Scout different from typing commands directly into a terminal?
Scout reads a plain language request, generates the correct CLI syntax, and runs it inside your workspace directory. You describe the outcome you want instead of writing the exact command yourself.
Does Scout run every command automatically?
No. Commands are split into auto-approve, prompt, and deny tiers. Only read-only or low-risk commands run without asking. Anything that changes state or touches the network pauses for approval, and destructive commands are blocked entirely.
Can I stop Scout from running shell commands?
Yes. Disabling the Shell category in Settings > Permissions removes shell access completely, including git, gh, curl, and PowerShell.
