Tools: read, write, explore

~15 min

Tools: read, write, explore

The agent is not typing into a terminal. It is calling tools — structured functions with inputs and outputs. Every action you see in the terminal panel is a tool call: bash to run a shell command, read_file to read a file, write_file to create one.

In this lesson you will watch the agent use its tools to explore data and organize a workspace.

Data lives in data/

Good practice: keep raw data in a data/ folder. Never modify source data — create derived files elsewhere. Your workspace already has data/trade_data.csv with international trade data for three countries.

Your task

Ask the agent to list the files in the workspace.

Ask it to preview the trade data — what columns does it have? How many rows?

Ask it to create a src/ directory for analysis scripts.

Watch the tool calls in the terminal. Notice how each one shows a name (like bash or list_csv) and its result.

Prompt to copy

List the files in the workspace, then preview data/trade_data.csv. How many rows and columns?

After it responds:

Prompt to copy

Create a src/ directory for analysis scripts.

What to notice

The agent calls bash with ls to list files — it is running a real shell command in a sandbox.

list_csv returns structured data (column names, types, sample rows) — not a screenshot, not a guess.

write_file and bash mkdir create real files you can see in the file tree on the right.

Every tool call is deterministic and inspectable. You can see exactly what the agent did.

Checkpoints

Workspace explored — The agent listed the files in the workspace.

Data previewed — The agent used list_csv to preview data/trade_data.csv.

Directory created — A src/ directory exists in the workspace.

Checkpoints

Workspace explored

Data previewed

Directory created

Was this lab useful?

claude-sonnet-4-6

Agent

Ask the agent to explore your data, write Stata code, or help with your analysis.

Enter to send · Shift+Enter for newline

Files

EditorSelect a file above