Skip to main content
Loading navigation
Back to challenges

Build a Claude Agents SDK Crypto Sanctions Investigator with Patronus AI

Cryptocurrency exchanges require precise multi-jurisdictional sanctions mapping to detect policy evasion. Build an autonomous investigation agent using Claude Agents SDK and evaluate policy alignment with Patronus AI to analyze entity transaction graphs with zero hallucinated compliance claims.

CybersecurityHosted by Vera
Sign in free to start

What you are building

Orchestrate Claude Agents SDK extended reasoning capabilities paired with Patronus AI evaluation frameworks to investigate complex crypto entity graphs.

How work is evaluated

Evaluation

Measures graph node traversal accuracy and Patronus AI hallucination evaluation score for compliance narratives.

Datasets

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

How submissions are scored

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

zero_hallucination_pass

Patronus AI evaluation score must indicate 0 hallucinated facts.

Binary check

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

Dimension 2

patronus_eval_score

Patronus model evaluation metric score. • target: 0.98 • range: 0.9-1

Binary check

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

Learning goals

  • Build complex investigation workflows using Claude Agents SDK

  • Integrate Patronus AI for real-time hallucination and policy evaluation

  • Trace high-velocity multi-hop crypto wallet transactions across blockchain networks

  • Generate immutable compliance audit trails adhering to regulatory standards

Resources

Dataset notes

Synthetic multi-hop cryptocurrency transaction graph dataset containing sanctioned clusters and mixing service interactions.

Challenge setup

Measures graph node traversal accuracy and Patronus AI hallucination evaluation score for compliance narratives.

Challenge input

JSON graph structure representing wallet address transfers

Claude Agents SDK

Provides extended context handling and logical deduction.

Patronus AI

Ensures factual precision and absence of agent hallucinations.

Evaluated output

JSON SAR summary detailing linked high-risk clusters

Evaluation checks
  • Patronus AI evaluation score must indicate 0 hallucinated facts.
  • Patronus model evaluation metric score. • target: 0.98 • range: 0.9-1
Configured evaluation evidence
  • Benchmark: Crypto Crime Graph Benchmark
  • Patronus Eval Score target: 0.98
  • 1 public reference case
Configured runtime
  • Python execution harness
  • Python sandbox (unavailable on Versalist)

Shows the first task, up to three tools, and up to three evidence items.

View technical recipe

Configured tools

Action Space
  • Claude Code · Optional
Observation
  • Claude Code · Optional
Reward / Eval
  • Patronus AI · Required
Policy Serving
  • Claude 4.1 Opus · Optional
Safety / Guardrails
  • Patronus AI · Required

Evaluation contract

  • zero_hallucination_pass · Weight 1
  • patronus_eval_score · Weight 1

Metrics

  • patronus_eval_score · Target 0.98

Benchmarks

  • Crypto Crime Graph Benchmark
  • security

Reference cases

  • 1 public reference case

Workflow tasks

  1. Task 1

    Input: JSON graph structure representing wallet address transfers

    Output: JSON SAR summary detailing linked high-risk clusters

    Execution harness: Python

Evaluation tools

  • Claude Agents SDK · Reasoning Agent Framework
  • Patronus AI · Evaluation Suite

Recipe freeze marker

No recipe freeze time is recorded.

Frequently Asked Questions about Build a Claude Agents SDK Crypto Sanctions Investigator with Patronus AI