SidekickSidekick

Agent

Agent Mode is Sidekick's primary AI assistant mode. It gives the AI full access to your project—reading files, writing code, searching your codebase, and running terminal commands—to help you build faster.

Overview

Agent Mode is the default mode when you start Sidekick. It's designed for everyday coding tasks.

AttributeValue
Icon∞ (Infinity)
ColorTheme color (blue)
Best forCode generation, debugging, refactoring, Q&A
ShortcutShift+Tab to cycle modes

What Agent Can Do

In Agent Mode, the AI has access to powerful tools to help you:

File Operations

  • Read Files — View any file in your project with optional line ranges
  • Write File — Create new files or completely rewrite existing ones
  • Edit File — Make targeted changes using find-and-replace
  • Scan Folder — List directory contents recursively or shallow
  • File Tree — Display your project structure

Search & Analysis

  • Search Files — Full-text pattern search across your codebase
  • Grep Files — Regular expression search with context
  • Memory — Save and retrieve project-specific notes and context

Development

  • Terminal — Execute shell commands (requires your approval)
  • MCP Tools — Access external tools via Model Context Protocol
  • Web Search — Search the internet for information
  • Subagents — Delegate to specialized analysis agents

How to Use Agent Mode

Activate Agent Mode

Agent Mode is active by default. If you're in another mode, click the Mode Selector in the input area and choose Agent, or press Shift+Tab to cycle through modes.

Ask your question or describe your task

Type naturally in the chat input. Be specific about what you need:

Help me refactor the authentication logic in @src/auth/login.ts
Find all places where we handle API errors and show me the patterns used
Create a new utility function that validates email addresses

Review AI actions

As the AI works, you'll see each tool it uses:

  • File reads show the content retrieved
  • Searches display matching results with context
  • Code changes appear as diffs you can review
  • Terminal commands require your approval before running

Approve or apply changes

  • Click Apply to accept code changes
  • Click Run to approve terminal commands
  • Click Skip to reject an action

Terminal Command Approval

For security, Agent Mode requires your approval before running terminal commands.

When the AI wants to run a command, you'll see:

  • The command text it wants to execute
  • Skip button — Reject this command
  • Run button — Execute the command
  • Run & Allowlist — Run and auto-approve this command in the future

Always review terminal commands before approving. Check the full command text (hover for details if truncated).

Allowlisting Commands

For commands you trust and use frequently (like npm test or git status), use Run & Allowlist. These commands will run automatically in the future without prompting.

Visual Indicators

While the AI is working, you'll see:

IndicatorMeaning
Spinning loaderTool is executing
Green checkmarkTool completed successfully
Red alertTool encountered an error
Expandable cardClick to see full tool input/output

Each tool call displays:

  • Tool name and icon
  • Input parameters (what the AI requested)
  • Results or output (when complete)
  • Timestamp

Best Practices

Be Specific

Instead of vague requests, provide context:

Fix my code
Fix the null pointer exception in @src/utils/parser.ts 
where we access user.profile without checking if user exists

Use File Mentions

Type @ to mention specific files or folders. This gives the AI direct context:

Review @src/components/Button.tsx for accessibility issues

Search for API calls in @src/services/

Enable Extended Thinking

For complex analysis tasks, toggle Extended Thinking in the input area. The AI will take more time but produce more thorough results.

When to Use Agent Mode

Code Generation

Writing new features, components, or utilities

Debugging

Finding and fixing bugs with AI assistance

Refactoring

Improving code structure and patterns

Code Review

Getting feedback on your implementations

Learning

Understanding unfamiliar codebases or concepts

Automation

Running build scripts, tests, and deployments