Build a Claude Agents SDK Reproductive Psychiatry Legal & Risk Inspector
High-risk psychiatric medications prescribed during postpartum conditions carry complex liability and adverse event reporting considerations. Build a Claude Agents SDK agent utilizing extended thinking capabilities to review psychiatric encounter records, flag liability exposure, and ensure protocol compliance.
What you are building
Construct an agent using Anthropic Claude Agents SDK with extended thinking to evaluate high-risk clinical records and legal compliance.
How work is evaluated
Evaluates extended thinking agent performance in detecting undocumented psychiatric risk factors.
Datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
flag_missing_consent
Ensures missing consent is captured in flagged_omissions list.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
risk_classification_precision
Precision of assigned liability levels across audited psychiatric chart benchmarks. • target: 0.92 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Learning goals
Build a Claude Agents SDK workflow with extended thinking enabled for deep clinical reasoning
Extract medication changes, black-box warning advisories, and consent documentation from notes
Assess risk scores for medico-legal liability in reproductive psychiatry cases
Output structured JSON compliance audits highlighting omitted clinical safeguards
Resources
25 de-identified synthetic psychiatric clinical notes detailing perinatal and postpartum treatment plans with variable documentation quality.
Challenge setup
Evaluates extended thinking agent performance in detecting undocumented psychiatric risk factors.
Challenge input
JSON containing encounter_note (string) and prescribed_meds (list of strings).
Claude Agents SDK
Official SDK with deep extended thinking for multi-step reasoning.
Evaluated output
JSON containing liability_level (string), flagged_omissions (list), and protocol_compliant (boolean).
- Ensures missing consent is captured in flagged_omissions list.
- Precision of assigned liability levels across audited psychiatric chart benchmarks. • target: 0.92 • range:...
- Benchmark: Medico-Legal Audit Benchmark
- Risk Classification Precision target: 0.92
- 1 public reference case
- 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
- ONNX Runtime · Required
- Retool · Optional
- Anthropic · Optional
Evaluation contract
- flag_missing_consent · Weight 1
- risk_classification_precision · Weight 1
Metrics
- risk_classification_precision · Target 0.92
Benchmarks
- Medico-Legal Audit Benchmark
- biotech
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON containing encounter_note (string) and prescribed_meds (list of strings).
Output: JSON containing liability_level (string), flagged_omissions (list), and protocol_compliant (boolean).
Execution harness: Python
Evaluation tools
- Claude Agents SDK · Agent Framework
Recipe freeze marker
No recipe freeze time is recorded.