Agent Foundry
All Problems

#5. Chain-of-Thought Activator

EasyPrompt Design

The Problem

Your math assistant agent is getting multi-step word problems wrong. When asked "Sarah buys 3 notebooks at $4 each, pays with $20, then buys 2 pens at $1.50 each — how much is left?", it jumps straight to a number and often gets it wrong. The model is capable of solving these problems correctly when it reasons through them, but the current system prompt tells it to "just state the final number," so it skips intermediate steps and makes arithmetic mistakes. Your job is to fix only the system prompt so the agent thinks step by step, shows its work, and arrives at the correct answer.

Examples

Example 1

User input: A store sells notebooks for $4 each. Sarah buys 3 notebooks and pays with a $20 bill. She then buys 2 pens at $1.50 each. How much money does she have left?

Current (bad) output: $6

Expected (good) output:

  • Step 1: Cost of 3 notebooks = 3 × $4 = $12
  • Step 2: Change from $20 = $20 − $12 = $8
  • Step 3: Cost of 2 pens = 2 × $1.50 = $3
  • Step 4: Money left = $8 − $3 = $5

Final answer: $5

Example 2

User input: A train travels 60 km/h for 2.5 hours, then 80 km/h for 1.5 hours. What is the total distance?

Current (bad) output: 250 km

Expected (good) output:

  • Step 1: Distance at 60 km/h for 2.5 h = 60 × 2.5 = 150 km
  • Step 2: Distance at 80 km/h for 1.5 h = 80 × 1.5 = 120 km
  • Step 3: Total distance = 150 + 120 = 270 km

Final answer: 270 km

Your Task

Update the system prompt (and only that) so the agent:

  • Breaks every problem into explicit numbered steps.
  • Shows each intermediate calculation clearly.
  • Arrives at the correct final answer after reasoning through each step.
  • Clearly labels the final answer separately from the reasoning.

Do not add tools, change model parameters, or add post-processing code.

Evaluation

Submissions are checked for the following:

  • Correct final answer: The agent produces the mathematically correct result.
  • Reasoning steps shown: The response contains explicit step-by-step reasoning before the final answer.
  • Intermediate calculations visible: Each sub-calculation is written out so the reader can follow and verify the logic.

Constraints

  • You may only modify the system prompt
  • The agent must show its reasoning before the final answer
  • No external calculator tools may be added
Starter Code
from langchain_openai import ChatOpenAI
from langchain_core.prompts import ChatPromptTemplate

llm = ChatOpenAI(model="gpt-4o-mini")

# BUG: This prompt encourages direct answers without showing reasoning steps
prompt = ChatPromptTemplate.from_messages([
    ("system", "You are a math assistant. Give the answer to the user's math problem. Be concise — just state the final number."),
    ("human", "{input}"),
])

chain = prompt | llm

# Test: multi-step word problem that requires intermediate calculations
result = chain.invoke({
    "input": "A store sells notebooks for $4 each. Sarah buys 3 notebooks and pays with a $20 bill. She then buys 2 pens at $1.50 each. How much money does she have left?"
})
print(result.content)
Open in Google Colab
Evaluation Criteria0/3