Claude Agents SDK Perinatal Psychiatry Safety Review
Perinatal mental health clinicians face complex diagnostic and liability landscapes requiring strict clinical decision support. Build a Claude Agents SDK application with extended thinking to evaluate electronic health records for postpartum risk indicators and generate liability risk reviews. Achieve 100 percent adherence to clinical safety guidelines.
What you are building
The core problem, expected build, and operating context for this challenge.
Develop a deep reasoning perinatal clinical review agent using Anthropic Claude Agents SDK with extended thinking to identify postpartum psychiatric risks.
How work is evaluated
Evaluates risk stratification accuracy, safety escalation compliance, and reasoning trace completeness.
Challenge datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.
safety_escalation_pass
Verifies that acute risk indicators trigger immediate escalation flag.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
risk_tier_accuracy
Accuracy of assigned risk tier compared to expert consensus. • target: 0.95 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Configure Claude Agents SDK agent with extended thinking budget for deep medical reasoning
Parse multi-visit EHR progression notes to identify subtle postpartum mood and psychosis red flags
Implement safety guardrails enforcing critical escalation protocols for acute risk flags
Generate structured clinical risk summaries with actionable intervention checklists
Reference links and supporting material
Synthetic clinical vignettes detailing perinatal psychiatric evaluations with annotated risk levels and safety escalation flags.
Challenge setup
Evaluates risk stratification accuracy, safety escalation compliance, and reasoning trace completeness.
Challenge input
JSON containing patient_history_text and visit_notes array
Claude Agents SDK
Required framework for deep reasoning agent implementations.
Claude 3.7 Sonnet
Supports extended thinking mode for complex differential diagnosis.
Evaluated output
JSON with risk_tier, acute_escalation_required, risk_factors, and clinical_plan
- Verifies that acute risk indicators trigger immediate escalation flag.
- Accuracy of assigned risk tier compared to expert consensus. • target: 0.95 • range: 0-1
- Benchmark: PerinatalSafety-Bench
- Risk Tier Accuracy target: 0.95
- 1 public reference case
- 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
- Lovable · Required
- Protecto · Optional
- DeepSeek R1 · Optional
Evaluation contract
- safety_escalation_pass · Weight 1
- risk_tier_accuracy · Weight 1
Metrics
- risk_tier_accuracy · Target 0.95
Benchmarks
- PerinatalSafety-Bench
- biotech
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON containing patient_history_text and visit_notes array
Output: JSON with risk_tier, acute_escalation_required, risk_factors, and clinical_plan
Execution harness: Python
Evaluation tools
- Claude Agents SDK · Agent core framework
- Claude 3.7 Sonnet · LLM Provider
Recipe freeze marker
No recipe freeze time is recorded.