Tool Permissions

An agent with unrestricted tools is a security incident waiting to happen. Permissions enforce least-privilege at the tool level.

Tool permissions are key cards — the intern's card opens the lobby; only the admin's opens the server room.

Scenario

A 'support agent' can search_tickets and refund_order (max $50); only an 'admin agent' can delete_user or modify_billing.

Solution

In Tool Calling & Function Calling, apply Tool Permissions to this scenario: A 'support agent' can search_tickets and refund_order (max $50); only an 'admin agent' can delete_user or modify_billing. Identify the inputs, run the technique, validate the output, and note one thing you would monitor in production.

Do this before moving to the next module — reading alone is not enough.

Open the Code Walkthrough below and run it locally. Change one parameter related to Tool Permissions (e.g. model, temperature, top_k, or tool name), observe the difference in output, and write 2–3 sentences explaining what changed.

Highlighted lines show where Tool Permissions happens in the code.

1# Tool Permissions — minimal example2from openai import OpenAI3
4client = OpenAI()  # create API client5
6# Ask the model to explain this topic7response = client.chat.completions.create(  # core API call for Tool Permissions8    model="gpt-4o-mini",9    messages=[10        {"role": "system", "content": "You explain tool permissions clearly."},11        {"role": "user", "content": f"What is tool permissions?"},12    ],13    temperature=0,14)15print(response.choices[0].message.content)  # show output for debugging

• Treating Tool Permissions as a black box without evaluation
• Ignoring cost and latency in production
• Skipping error handling for tool permissions

Quick recap — the most important points from this module.