Claude Agents SDK Situational Awareness Diagnostic with Qwen 3
Network switch failures in process plants can disrupt control logic traffic and blind operators to critical safety alarms. Build an diagnostic agent using Claude Agents SDK integrated with Qwen 3 reasoning models to analyze network failure logs and operator alarm histories. Success is measured by correctly isolating the switch failure root cause within 30 seconds of simulated stream processing.
What you are building
Deploy a Claude Agents SDK agent paired with Qwen 3 inference to perform root cause analysis on network switch disruptions impacting safety-critical logic.
How work is evaluated
Evaluates root cause detection and operator guidance accuracy during network switch outages.
Datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
correct_root_cause_node
Checks if root cause component is correctly identified.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
rca_precision
Precision of identified root cause and recommended remediation steps. • target: 0.95 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Learning goals
Implement a Claude Agents SDK agent framework setup for diagnostic workflows.
Integrate Qwen 3 model reasoning for complex physical system state analysis.
Correlate network packet drop logs with PLC lost-heartbeat events.
Provide step-by-step situational awareness recovery actions for plant operators.
Resources
Combined network packet capture traces (PCAP), SNMP traps, and SCADA alarm logs during simulated industrial switch outages.
Challenge setup
Evaluates root cause detection and operator guidance accuracy during network switch outages.
Challenge input
JSON object with snmp_traps and scada_alarms arrays
Claude Agents SDK
Required framework for building multi-turn agent workflows.
Qwen 3
Utilized for complex multi-source logical reasoning and fault isolation.
Evaluated output
JSON object with root_cause_component, primary_impact, and recommended_action
- Checks if root cause component is correctly identified.
- Precision of identified root cause and recommended remediation steps. • target: 0.95 • range: 0-1
- Benchmark: OTNetworkRCA-2026
- Rca Precision target: 0.95
- 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
- Claude Code · Optional
- Claude Code · Optional
- Qwen 3 · Required
- Claude 4.1 Opus · Optional
Evaluation contract
- correct_root_cause_node · Weight 1
- rca_precision · Weight 1
Metrics
- rca_precision · Target 0.95
Benchmarks
- OTNetworkRCA-2026
- security
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON object with snmp_traps and scada_alarms arrays
Output: JSON object with root_cause_component, primary_impact, and recommended_action
Execution harness: Python
Evaluation tools
- Claude Agents SDK · Agent Framework
- Qwen 3 · Models
Recipe freeze marker
No recipe freeze time is recorded.