Dispatch Tool Calls Without Crashing the Agenteasy

Dispatch Tool Calls Without Crashing the Agent

easy Build an Agent 10–15 min

Standard library only · runs in your browser

Tools fail all the time: a city with no data, a divide by zero, a tool name the model invented. A production dispatcher turns every one of those into an observation the model can read. It never raises one that kills the run.

Task

Implement dispatch(call), where call is {"tool": name, "arguments": {...}} (what stage 1 returns). Look the tool up in the provided TOOLS registry and always return a string:

Unknown tool: error: unknown tool 'NAME'. available tools: add, divide, get_weather, lookup_doc (names sorted, comma-separated)
Invalid arguments (use your stage-2 validate_arguments, provided below): error: invalid arguments for 'NAME': followed by the errors joined with ;
The tool raises: error: NAME raised ExceptionType: message
Success: the result itself if it's a string, otherwise json.dumps(result)

Function Signature

dispatch(call: dict) -> str

Examples

Example 1: Numeric result is serialized
Input: call = {'tool': 'add', 'arguments': {'a': 2, 'b': 3}}
Output: '5'

Example 2: String result is returned as-is
Input: call = {'tool': 'get_weather', 'arguments': {'city': 'Paris'}}
Output: '18C, light rain'

Example 3: Hallucinated tool name
Input: call = {'tool': 'search_web', 'arguments': {'query': 'x'}}
Output: 'error: unknown tool \'search_web\'. available tools: add, divide, get_weather, lookup_doc'

Example 4: A tool that raises
Input: call = {'tool': 'divide', 'arguments': {'a': 1, 'b': 0}}
Output: 'error: divide raised ZeroDivisionError: division by zero'

Constraints

Never let an exception escape `dispatch`.
Every return value is a `str`: the model only reads text.
`TOOLS[name]` is `{"fn": callable, "schema": dict}`, so call it as `fn(**arguments)`.

Python3Saved ✓

import json

import math

import re

def dispatch(call):

"""Run call["tool"] with call["arguments"]; always return a string observation."""

# Your code here

pass

# =============================================================================

# PROVIDED BELOW -- the sandbox and your earlier stages. Read it, don't edit it.

# =============================================================================

# ---------------------------------------------------------------------------

# The sandbox world your agent lives in (provided -- you don't edit this).

# Four deterministic tools, each with a JSON-Schema-style argument spec,

# exactly the shape you'd hand a real model's tool-calling API.

# ---------------------------------------------------------------------------

WEATHER = {"paris": "18C, light rain", "tokyo": "24C, clear", "lima": "16C, overcast"}

DOCS = {

"refund-policy": "Refunds are accepted within 30 days with a receipt.",

"shipping": "Standard shipping takes 3-5 business days.",

}

def add(a, b):

return a + b

def divide(a, b):

return a / b

def get_weather(city):

key = city.strip().lower()

Case 1: Numeric result is serialized

Input: call = {'tool': 'add', 'arguments': {'a': 2, 'b': 3}}

Expected: '5'

Case 2: String result is returned as-is

Input: call = {'tool': 'get_weather', 'arguments': {'city': 'Paris'}}

Expected: '18C, light rain'

Case 3: Hallucinated tool name

Input: call = {'tool': 'search_web', 'arguments': {'query': 'x'}}

Expected: 'error: unknown tool \'search_web\'. available tools: add, divide, get_weather, lookup_doc'

Case 4: A tool that raises

Input: call = {'tool': 'divide', 'arguments': {'a': 1, 'b': 0}}

Expected: 'error: divide raised ZeroDivisionError: division by zero'