agent tools
The agent has access to specific tools that enable it to work with your codebase. Understanding what the agent can do helps you use it effectively.
available tools
file read
What it does: Reads the contents of files in your project.
How the agent uses it:
- Examines code to understand structure
- Finds relevant functions or classes
- Checks existing implementations before making changes
Example:
show me the authentication logic
The agent will search for and read auth-related files.
code search
What it does: Searches your codebase for specific patterns, functions, or text.
How the agent uses it:
- Finds where a function is defined
- Locates all uses of a particular API
- Searches for specific patterns across files
Example:
find all places where we call the payment api
The agent searches the codebase and returns matching locations.
file edit
What it does: Proposes modifications to code files.
How the agent uses it:
- Adds new functions or classes
- Modifies existing code
- Refactors across multiple files
You approve before changes are applied: The agent shows you a diff of proposed changes. You review and approve before anything is written.
Example:
add error handling to the checkout function
The agent reads the function, proposes changes with try/catch, and waits for your approval.
terminal execution
What it does: Runs commands in the built-in terminal.
How the agent uses it:
- Runs tests to verify changes work
- Installs dependencies
- Checks build output
- Executes Git commands
You approve before commands run: The agent shows you the command it wants to run. You approve before execution.
Example:
run the tests for the auth module
The agent determines the correct test command and executes it after your approval.
what the agent cannot do
Understanding limitations helps set appropriate expectations:
no web access
The agent cannot browse the internet, fetch external documentation, or access online services directly.
If you need external information, you can provide it in the conversation.
no direct api calls
The agent cannot make HTTP requests to external APIs unless they're accessible through Model Context Protocol (MCP) servers.
no file system operations outside the project
The agent works within your project directory. It cannot access files elsewhere on your system.
no magic
The agent needs clear instructions. It cannot read your mind or infer unstated requirements.
choosing the right tool for the task
good for the agent
Multi-file changes:
update all our api calls to use the new error handling pattern
Exploratory tasks:
explain how authentication works in this codebase
Context-heavy modifications:
refactor this component to use our new state management approach
faster to do manually
Single-line changes when you know exactly what to type
Simple typo fixes in one location
Rapid experimentation where you're faster than explaining
combining tools
The agent often uses multiple tools in sequence:
- Search for relevant files
- Read those files to understand context
- Propose edits with the necessary changes
- Run tests to verify the changes work
You see this workflow in the conversation as the agent progresses through your request.
agent capabilities summary
| tool | what it does | requires approval |
|---|---|---|
| file read | reads code from your project | no |
| code search | finds patterns across files | no |
| file edit | modifies code | yes |
| terminal | runs commands | yes |
related pages
- working with the agent - learn to prompt effectively
- terminal - using the built-in terminal
- glossary - key terms explained