Day 1 · Block 4

E04 - Tool-calling mini-agent

• Take a look at the LangChain documentation for tool use
• Think about your work. What kind of "tool" would be handy there?
• Take a look at https://agentskills.io/home and create a mockup of an agent skill that would be useful for your work.
• Run `python exercises/04/tool_agent.py`.
• Modify one `TODO-STUDENT` item (region map or question format) and rerun.
• Keep and report the trace showing both tool calls and plausibility check.

Inputs

• exercises/04/tool_agent.py
• Two tools: label lookup + average
• One multi-step question requiring both tools

Deliverable

tool trace with two tool calls + plausibility check + one TODO change note.

Target

Take a look at the LangChain documentation for tool use

Submission

• Use the live launcher: Live Exercises
• Direct prefilled form: E04 submission link

Checklist

• Make one explicit design decision.
• Include one verification check.
• State one limitation or risk.

Common failure modes

• Agent answers directly without using both tools.
• No no-auth fallback when external tool setup fails.

Extension task (optional)

Add graceful fallback for a forced tool failure and show fallback trace.