Skip to main content
Back to challenges
Challenge

Build a High-Trust Public Service Agent with Claude Agents SDK and Human-in-the-Loop

Build a high-trust public service automation workflow with human-in-the-loop review gates for Singapore Smart Nation applications. Powered by Claude Agents SDK, your system will process citizen benefits requests, evaluate eligibility rules, and route exceptions for human approval.

Workflow AutomationHosted 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 high-trust public service agent with Claude Agents SDK featuring explicit human-in-the-loop governance for civic benefit distributions.

Delivery guide

How work is evaluated

Evaluation

Evaluate decision correctness and human approval routing accuracy for civic benefit applications.

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

hitl_trigger_correctness

Correctly flags special circumstances for human review

Binary check

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

Dimension 2

eligibility_precision

Precision of automated benefit approvals • target: 0.96 • 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 Claude Agents SDK workflows with strict safety constraints and audit logs

  • Build human-in-the-loop approval handoffs for edge-case eligibility scenarios

  • Ensure compliance with Singapore Smart Nation ethical AI governance standards

  • Design zero-hallucination policy compliance tools for government benefit distribution

Resources and assets

Reference links and supporting material

Dataset notes

Synthetic citizen benefit application dossiers containing income levels, property valuations, and policy edge cases.

Challenge setup

Evaluate decision correctness and human approval routing accuracy for civic benefit applications.

Challenge input

JSON with applicant_id, household_income, property_pree_val, special_circumstances

Claude Agents SDK

Anthropic SDK supporting extended reasoning and structured tool interaction.

Evaluated output

JSON with eligible (bool), requires_human_review (bool), decision_reason

Evaluation checks
  • Correctly flags special circumstances for human review
  • Precision of automated benefit approvals • target: 0.96 • range: 0-1
Configured evaluation evidence
  • Benchmark: GovTech_Public_Trust_Bench
  • Eligibility Precision target: 0.96
  • 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
  • Ludwig · Optional
  • Embedchain · Optional
Policy Serving
  • OpenRouter · Required
Orchestration
  • OpenRouter · Required

Evaluation contract

  • hitl_trigger_correctness · Weight 1
  • eligibility_precision · Weight 1

Metrics

  • eligibility_precision · Target 0.96

Benchmarks

  • GovTech_Public_Trust_Bench
  • general

Reference cases

  • 1 public reference case

Workflow tasks

  1. Task 1

    Input: JSON with applicant_id, household_income, property_pree_val, special_circumstances

    Output: JSON with eligible (bool), requires_human_review (bool), decision_reason

    Execution harness: Python

Evaluation tools

  • Claude Agents SDK · Agent Framework

Recipe freeze marker

No recipe freeze time is recorded.

Frequently Asked Questions about Build a High-Trust Public Service Agent with Claude Agents SDK and Human-in-the-Loop