Supply Chain Resilience Swarm Agents
Recent supply chain attacks on tools like LiteLLM highlight the need for automated incident response. In this challenge, you will use CrewAI to orchestrate a team of agents that monitor data pipelines and react to security breaches. You will utilize Protecto for data privacy masking within agent communications and Agency for real-time observability of the agent swarm. The system will leverage the GPT-5.3-Codex model for its core reasoning engine, ensuring that incidents are handled with high precision and logical consistency.
What you are building
The core problem, expected build, and operating context for this challenge.
Build a multi-agent swarm with CrewAI and GPT-5.3-Codex to monitor and respond to supply chain security incidents.
How work is evaluated
Evaluation centers on the swarm's ability to triage a mock supply chain incident and correctly delegate tasks.
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.
PII Masking Check
Agent communications must not contain unmasked sensitive data.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Delegation Efficiency
Percentage of tasks correctly routed to the specialized agent on the first attempt • target: 90 • range: 0-100
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Master CrewAI role-playing agent teams with specialized agents for incident triage, forensic analysis, and containment
Integrate Protecto to ensure that sensitive company data is masked before being processed by the GPT-5.3-Codex reasoning engine
Deploy Agentic Dev frameworks to accelerate the creation of custom agent tools for pipeline management
Implement Agency (Observability) to track agent thoughts and tool executions in a production-like environment
Leverage Ellipsis to provide an interface for human operators to approve agent-suggested containment actions
Design complex task dependencies in CrewAI that ensure forensic evidence is gathered before remediation begins
Reference links and supporting material
Sample data for 1 tasks
Challenge setup
Evaluation centers on the swarm's ability to triage a mock supply chain incident and correctly delegate tasks.
Challenge input
Log stream containing a mix of normal and malicious activity
CrewAI
Framework for orchestrating
Protecto
AI data privacy
Evaluated output
Categorized incident report with severity score
- Agent communications must not contain unmasked sensitive data.
- Percentage of tasks correctly routed to the specialized agent on the first attempt • target: 90 • range: 0-100
- Delegation Efficiency target: 90
- 1 public reference case
- Docker execution harness
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
- CrewAI · Required
- crewAI · Optional
- Protecto · Optional
Evaluation contract
- PII Masking Check · Weight 1
- Delegation Efficiency · Weight 1
Metrics
- Delegation Efficiency · Target 90
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: Log stream containing a mix of normal and malicious activity
Output: Categorized incident report with severity score
Execution harness: Docker
Recipe freeze marker
No recipe freeze time is recorded.