YOLO mode
"chat.tools.global.autoApprove": true (old name chat.tools.autoApprove): every tool in every workspace runs with no prompt.
AI coding agents · VS Code
A few chat.* settings decide which tools and terminal commands run without a prompt, and a trusted folder lets its MCP servers start on their own. VS Code itself calls terminal auto-approval "not a security boundary".
settings.json −risky+fixed−"chat.tools.global.autoApprove": true,−"chat.tools.terminal.autoApprove": { "/.*/": true }+"chat.tools.global.autoApprove": false,+"chat.tools.terminal.autoApprove": { "git status": true }
User settings apply everywhere; a project's .vscode files override them.
| Scope | File | What's in it |
|---|---|---|
| You | ~/Library/Application Support/Code/User/settings.json (macOS)%APPDATA%\Code\User\settings.json (Windows)~/.config/Code/User/settings.json (Linux) | The chat.* approval settings. |
| This project | .vscode/settings.json | Overrides yours; the terminal rules are merged. |
| MCP servers | Code/User/mcp.json.vscode/mcp.json.mcp.json | Servers it starts. Repo servers follow Workspace Trust. |
| Extensions | ~/.vscode/extensions | What you installed. |
| Admin policy | /etc/vscode/policy.json (Linux)a configuration profile (macOS)Group Policy (Windows) | Wins over every other setting. |
All in settings.json.
| Setting | What it does |
|---|---|
chat.permissions.default | default (Manual), autoApprove (Allow all: every tool call without confirmation) or autopilot (approves everything and answers its own questions). |
chat.tools.global.autoApprove | true approves every tool in every workspace. VS Code's own name for it: YOLO mode. /yolo does the same for one chat. |
chat.tools.terminal.autoApprove | Rules per command or /regex/: true runs it without asking, false asks. A false rule asks; it doesn't block. |
chat.tools.terminal.ignoreDefaultAutoApproveRules | true drops the built-in rules that ask before rm, curl, chmod and others. |
"chat.tools.global.autoApprove": true (old name chat.tools.autoApprove): every tool in every workspace runs with no prompt.
"chat.tools.terminal.autoApprove": { "/.*/": true }: every command runs without asking.
"chat.permissions.default": "autoApprove" or "autopilot": every new chat starts there.
Once you trust the folder, .vscode/mcp.json and .mcp.json servers can start without a separate prompt, and by default when you send a chat message.
"security.workspace.trust.enabled": false turns off Restricted Mode, which keeps agents and repo MCP servers out of unfamiliar folders.
"github.copilot.chat.claudeAgent.allowDangerouslySkipPermissions": true bypasses every permission check for the Claude agent in VS Code.
In your user settings.json, and remove any catch-all rule such as "/.*/": true from chat.tools.terminal.autoApprove.
{
"chat.permissions.default": "default",
"chat.tools.global.autoApprove": false,
"chat.tools.terminal.ignoreDefaultAutoApproveRules": false,
"chat.mcp.autostart": "never",
"security.workspace.trust.enabled": true
} New chats start in Manual, no tool is approved everywhere, the built-in rules for rm and curl stay on, no MCP server starts just because you sent a message, and unfamiliar folders open in Restricted Mode. Stricter: "chat.tools.terminal.enableAutoApprove": false.
For VS Code, NeuroShield reads your user settings.json, each project's .vscode/settings.json, mcp.json files and extensions: YOLO mode, catch-all terminal rules (Fix It for Me removes the rule, with Undo), and the MCP checks for keys and downloads.
It reads settings files and never runs your agents, their MCP servers or their scripts.
The check runs locally. No account, no upload, no AI inside: fixed, published rules.
Where a safe fix exists, Fix It for Me makes it after you confirm, with a backup and Undo.
The setting chat.tools.global.autoApprove: it disables manual approval for every tool in every workspace. /yolo turns it on for one chat.
No. A false rule asks first; it doesn't block. It does win over any true rule.
Not in Restricted Mode. Once you trust the folder they can, and by default VS Code starts new or changed servers when you send a chat message.
The check is free: how many agents you have and what they can do. Details and fixes come with Shields.