Auto-approve

Before Bob runs a tool, it checks whether the action is pre-approved for the active task. If pre-approved, the action runs immediately. If not, an approval card opens in the chat interface.

Approval prompts appear one at a time so you have a clear, deliberate view of each action before it runs. Large details (such as a full file diff or command output) collapse into a preview frame by default. Click to expand them when you want to inspect full details.

Auto-approval in Bob IDE combines four distinct controls across different scopes:

When you auto-approve an action, Bob will complete the respective action without asking for permission.

When you turn on the Execute tool group, Bob does not automatically run every terminal command. Auto-approval for commands requires matching Allowed Commands rules.

Under Settings → Execute, you can specify command prefix patterns:

When Command Security Verification is turned on under Settings → Execute, Bob performs security heuristics on commands before running them. If a command involves high-risk patterns (such as potential credential exfiltration, destructive deletions, or elevated privilege requests), Bob flags the command for manual confirmation even if the binary is on your Allowed Commands list.

When a command requires confirmation, the approval card provides the following options:

The MCP tool group works differently from standard groups:

Auto-approve is most beneficial in these scenarios:

Each action controls a specific workflow that Bob can complete.

Bob accesses and reads files in your project without asking for permission.

This includes viewing file contents, listing directory structures, and searching through your codebase. This permission involves information gathering rather than making changes to your system.

Bob creates, modifies, and deletes files without asking for permission.

Bob can apply code changes, refactor existing code, and manage file structure automatically. This directly modifies your codebase and involves high risk.

Bob can run terminal commands automatically without asking permission. This includes running build scripts, package managers, and other command-line tools.

Commands can complete system-level operations, and possibly run harmful processes. This action involves high risk.

Bob can use configured Model Context Protocol (MCP) servers without asking permission.

MCP servers provide additional tools and resources that extend Bob's capabilities. The risk level depends on what the specific MCP servers can access and modify.

Bob can automatically activate skills you have defined without asking for permission.

Skills are reusable instruction sets that teach Bob specialized workflows and tasks. When enabled, Bob will load and follow skill instructions based on your request and the skill's description. The risk level depends on what actions the skill instructions direct Bob to perform, but skills operate within Bob's existing tool constraints.

For more information, see Skills.

Bob can automatically update the todo list without asking for permission.

This allows Bob to track progress and manage task completion as work proceeds. This permission relates to task management rather than system access.

Subtasks are separate task instances that Bob creates to break down complex work into manageable pieces. Bob can create and complete these subtasks automatically without asking for confirmation.

This permission relates to workflow organization rather than system access.

Bob can automatically spawn subagents to handle focused, self-contained tasks without asking for permission.

Subagents run in their own isolated context and return a summary when done. This permission relates to workflow delegation rather than direct system access.

For more information, see Subagents.

Bob can change between different modes automatically based on task requirements.

This action involves workflow organization and does not affect system access.

Follow these guidelines to use auto-approving actions safely: