Banking Infrastructure Threat Defense with Claude Agents SDK and Protect AI
Advanced persistent threat campaigns targeting core banking APIs demand automated threat hunting and rapid vulnerability mitigation. You will build an autonomous security agent that inspects API traffic logs for zero-day exploit patterns and automatically generates firewall patch configurations, successfully mitigating attack vectors under simulated adversary conditions.
What you are building
Leverage Claude Agents SDK combined with Protect AI security guardrails to build an autonomous threat hunting and defense pipeline for core banking APIs.
How work is evaluated
Evaluates agent ability to classify malicious traffic patterns and generate syntactically correct WAF rules.
Datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
waf_syntax_valid
Checks whether generated WAF rule parses cleanly without errors
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
mitigation_rate
Percentage of malicious attacks correctly mitigated • target: 0.98 • range: 0.9-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Learning goals
Deploy Claude Agents SDK with computer use and extended thinking capabilities
Integrate Protect AI runtime security scanners to inspect LLM inputs and responses
Analyze network traffic logs to detect advanced persistent threat (APT) indicators
Synthesize automated Web Application Firewall (WAF) rule sets to block exploits
Resources
Synthetic core banking API log stream containing normal transaction traffic mixed with sophisticated nation-state attack vectors.
Challenge setup
Evaluates agent ability to classify malicious traffic patterns and generate syntactically correct WAF rules.
Challenge input
JSON log snippet detailing malicious payload
Claude Agents SDK
Primary framework powering agent reasoning and extended thinking cycles.
Protect AI
Provides runtime guardrails and prompt safety scanning.
Evaluated output
JSON containing threat_type, detected_signature, and generated_waf_rule
- Checks whether generated WAF rule parses cleanly without errors
- Percentage of malicious attacks correctly mitigated • target: 0.98 • range: 0.9-1
- Benchmark: CyberDef-Bench
- Mitigation Rate target: 98%
- 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
- Claude 4.1 Opus · Optional
- Protect AI · Required
Evaluation contract
- waf_syntax_valid · Weight 1
- mitigation_rate · Weight 1
Metrics
- mitigation_rate · Target 0.98
Benchmarks
- CyberDef-Bench
- security
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON log snippet detailing malicious payload
Output: JSON containing threat_type, detected_signature, and generated_waf_rule
Execution harness: Python
Evaluation tools
- Claude Agents SDK · agent_framework
- Protect AI · ai_engineering_tooling
Recipe freeze marker
No recipe freeze time is recorded.