Skip to main content
Back to challenges
Challenge

Autonomous Service Dispute Resolver using Claude Agents SDK and Fixie

Urban Company's jump in revenue alongside a shift 'into the red' highlights the critical need for operational efficiency, especially in managing service quality and disputes. This challenge focuses on building an autonomous Service Dispute Resolver using the Claude Agents SDK. You will create an agent that handles customer complaints about home services (e.g., plumbing, cleaning) by analyzing technician logs, customer photos of the 'issue,' and service history. The agent must utilize Claude’s 'extended thinking' (Chain of Thought) to decide whether to issue a refund, schedule a free rework, or deny the claim based on Urban Company's service guidelines. You will use Fixie to build the conversational interface and manage the 'side-channel' tools that allow the agent to look up technician ratings and past customer behavior in real-time.

Business OperationsHosted by Vera
Sign in free to start
Challenge brief

What you are building

The core problem, expected build, and operating context for this challenge.

Develop a complex reasoning agent using Claude Agents SDK and Fixie to automate service quality dispute resolutions in the home-services sector.

Delivery guide

How work is evaluated

Evaluation

Evaluation is based on 'Fairness' and 'Accuracy' scores—comparing the agent's resolution against a set of expert-moderated dispute outcomes.

Datasets

Challenge datasets

Public datasets, plus the private datasets you have access to.

Evaluation rubric

How submissions are scored

These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.

Dimensions
2 scoring checks
Binary
2 pass or fail dimensions
Ordinal
0 scaled dimensions
Dimension 1

Policy Adherence

Agent must cite a valid clause from the provided service manual.

Binary check

This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.

Dimension 2

Resolution Consistency

Variance in decisions across 5 identical runs. • target: 0.95 • range: 0-1

Binary check

This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.

Learning goals

What you should walk away with

  • Implement the Claude Agents SDK for complex multi-step reasoning workflows

  • Leverage Anthropic's 'Extended Thinking' capabilities for fair dispute resolution

  • Build interactive agent interfaces using Fixie to bridge LLMs with enterprise data

  • Design policy-compliant agents that adhere to strict service-level agreements (SLAs)

Resources and assets

Reference links and supporting material

Dataset notes

A collection of 500 anonymized service tickets including text logs, simulated image metadata, and policy documents.

Challenge setup

Evaluation is based on 'Fairness' and 'Accuracy' scores—comparing the agent's resolution against a set of expert-moderated dispute outcomes.

Challenge input

JSON with customer_complaint, technician_notes, and photo_analysis_metadata

Claude Agents SDK

Allows native integration of Claude's thinking and tool use.

Fixie

Used to create the conversational wrapper and manage API sidecars.

Claude 3.5 Sonnet

Top-tier reasoning and tool use capabilities.

Evaluated output

JSON with decision, justification, and policy_citation

Evaluation checks
  • Agent must cite a valid clause from the provided service manual.
  • Variance in decisions across 5 identical runs. • target: 0.95 • range: 0-1
Configured evaluation evidence
  • Benchmark: CS-Agent-Reasoning
  • Resolution Consistency target: 0.95
  • 1 public reference case
Configured runtime
  • Python execution harness
  • Python sandbox (unavailable on Versalist)

Summary shows the first task, up to three tools, and up to three evidence items. Open the technical recipe to view all configured items.

View technical recipe

Configured tools

Action Space
  • Fixie · Required
  • Aim · Optional
  • Anthropic Claude · Optional

Evaluation contract

  • Policy Adherence · Weight 1
  • Resolution Consistency · Weight 1

Metrics

  • Resolution Consistency · Target 0.95

Benchmarks

  • CS-Agent-Reasoning
  • general

Reference cases

  • 1 public reference case

Workflow tasks

  1. Task 1

    Input: JSON with customer_complaint, technician_notes, and photo_analysis_metadata

    Output: JSON with decision, justification, and policy_citation

    Execution harness: Python

Evaluation tools

  • Claude Agents SDK · Agent Framework
  • Fixie · Interface & Tooling
  • Claude 3.5 Sonnet · Model

Recipe freeze marker

No recipe freeze time is recorded.

Frequently Asked Questions about Autonomous Service Dispute Resolver using Claude Agents SDK and Fixie