Coding Question 45 minutesHard

Stock Research Agent

Repair a tool-using research agent, then make its execution loop cheaper and faster.

Question

Repair the stock-research implementation in research_agent.py. It must answer the requested historical-price and YTD-return question while following the model's multi-tool message protocol exactly and efficiently.

Function parameters

price_api: For get_stock_price: exposes close_on(symbol, date); the requested date must be forwarded unchanged.
first_close / latest_close: For get_ytd_return: numeric first trading close of the year and latest close.
client: For run_agent: model client called as create(system=SYSTEM_PROMPT, messages=messages).
tools: For run_agent: dictionary from requested tool name to a callable accepting the supplied input fields.
user_request: For run_agent: original question placed in the initial user message.

Inputs and API responses

Messages API

Entity	Fields
Response	stop_reason, content[]
ToolUseBlock	id, type, name, input
ToolResultBlock	type, tool_use_id, content

One assistant response may request several tools.
Stop only when stop_reason is end_turn.

Market data

Entity	Fields
PriceAPI	close_on(symbol, date)
Tools	get_stock_price, get_ytd_return

Historical date arguments are meaningful.
Return percentages rounded to two decimals.

Agent contract

tool_result.tool_use_id must match the originating ToolUseBlock.id
Execute all tool calls from one response before the next model call
The loop continues until the model returns end_turn

Requirements

get_stock_price must use the requested date.
get_ytd_return must calculate (latest - first) / first * 100.
Append the complete assistant response before returning tool results.
Return one tool_result for every tool_use block, preserving order.
Use tool_use_id, not id, in tool-result dictionaries.
Do not call the model once per tool; batch results into the next turn.
A program can run without exceptions and still violate the Messages API protocol or silently answer a different financial question.

Editable

SYSTEM_PROMPT = """You are a stock research agent. Use the available tools and answer with evidence."""

def get_stock_price(price_api, symbol, date):
    # BUG: this ignores the requested historical date.
    return price_api.latest_close(symbol)

def get_ytd_return(first_close, latest_close):
    # BUG: the direction and denominator are wrong.
    return round((first_close - latest_close) / latest_close * 100, 2)

def run_agent(client, tools, user_request):
    messages = [{"role": "user", "content": user_request}]
    response = client.create(system=SYSTEM_PROMPT, messages=messages)

# BUG: this handles only one round of tool use.
    if response["stop_reason"] == "tool_use":
        tool_results = []
        for block in response["content"]:
            if block["type"] != "tool_use":
                continue
            result = tools[block["name"]](**block["input"])
            tool_results.append({
                "type": "tool_result",
                "id": block["id"],
                "content": str(result),
            })
            messages.append({"role": "assistant", "content": response["content"]})
            messages.append({"role": "user", "content": tool_results})
            response = client.create(system=SYSTEM_PROMPT, messages=messages)

return response

Test output

Run the challenge to validate the agent protocol, calculations, and model-turn count.