CSEC 602: Noctua — Advanced Agentic Security Engineering

Semester 2 Syllabus

Term: Semester 2, 2026 Credits: 3 units Duration: 16 weeks Prerequisite: CSEC 601 Delivery Mode: Hybrid (lecture + hands-on labs) Lab-to-Theory Ratio: 70% hands-on, 30% theory

Course Information

Course Overview

By the end of Semester 2, you will be able to: assess a multi-agent system for security risks and architectural gaps; harden agents against adversarial manipulation and real-world attack patterns; and deploy observable, governed, production-ready systems. These are the skills that produce value in a company engagement — not just technically, but operationally.

Building on CSEC 601 foundations, this semester scales from individual agents to orchestrated systems, from building to adversarial testing, and from prototype to production. The frameworks that enable this — Claude Agent SDK, CrewAI, LangGraph, and MCP — are taught as tools in service of those outcomes, not as ends in themselves.

Key Terminology

Collaborative Critical Thinking (CCT): The systematic application of structured reasoning to design robust agentic systems, considering agent interactions, failure modes, security boundaries, and ethical implications. This is distinct from "prompting" and represents deep architectural and design thinking.
Context Engineering: The intentional design and management of information context, instruction clarity, and tool availability to shape agent behavior. This is distinct from "prompt engineering" and encompasses the broader challenge of engineering systems, not just text prompts.

Lab Philosophy

Semester 2 labs maximize the capabilities of Claude Max subscriptions and the cutting-edge agentic toolstack:

Claude Code and Agent SDK: Students leverage Claude's IDE, real-time code execution, worktrees for branch isolation, and the Agent SDK's subagent capabilities for orchestrating specialized agent teams.
Multi-Vendor Perspective: Beyond Claude's ecosystem, students gain hands-on experience with CrewAI, LangGraph, and AutoGen to understand architectural trade-offs across frameworks.
Autonomous Rapid Prototyping: By Semester 2, students are expected to be proficient with the agentic toolstack and are expected to prototype complex systems quickly, iterating on design through building rather than extensive planning.
Production Focus: Labs shift from "proof of concept" to "production-ready," incorporating observability, error handling, deployment strategies, and operational considerations.

Learning Objectives

By the end of this course, students will be able to:

Design and implement multi-agent architectures that orchestrate specialized agents for complex cybersecurity tasks, comparing supervisor, hierarchical, debate, and swarm patterns.
Evaluate and select appropriate agentic frameworks (Claude Agent SDK, CrewAI, LangGraph) based on security requirements, workflow characteristics, and operational constraints.
Conduct Collaborative Critical Thinking analysis to design agent systems that are secure, efficient, and resilient to adversarial manipulation.
Execute red team and blue team operations targeting AI agents, including prompt injection, goal hijacking, tool misuse, and memory poisoning attacks.
Implement guardrails, hardening, and defense strategies to protect AI agents from adversarial manipulation while maintaining operational effectiveness.
Apply MITRE ATLAS threat modeling to identify, prioritize, and mitigate AI-specific cybersecurity risks in autonomous systems.
Design and implement supply chain security for AI systems, including model provenance, dependency management, and training data integrity verification.
Govern non-human identities (NHI) in multi-agent environments through authentication, authorization, and continuous audit mechanisms.
Implement observability and monitoring for agentic systems in production using OpenTelemetry, token tracking, cost management, and anomaly detection.
Deploy agentic security systems to production environments with CI/CD pipelines, containerization, canary deployments, and operational runbooks.
Integrate ethical reasoning and responsible AI principles into the design and operation of autonomous security systems.
Build, test, and defend real-world agentic security solutions that demonstrate mastery of the full system lifecycle from design to operation.

Course Structure

Weekly Format

Each week consists of two class sessions:

Day 1 — Theory & Foundations: Conceptual foundations, historical context, case studies, guided discussions, and key concept development
Day 2 — Hands-On Lab: Practical building using Claude Code and the course toolstack, with immediate prototyping, iteration, and deliverable completion

Time Commitment

Day 1 (Theory): 2 hours per week
Day 2 (Lab): 4-6 hours per week
Independent work: 4-6 hours per week
Total: 10-14 hours per week, 160-224 hours over 16 weeks

Unit Organization

Unit 5 (Weeks 1-4): Multi-Agent Orchestration for Security
Unit 6 (Weeks 5-8): AI Attacker vs. AI Defender
Unit 7 (Weeks 9-12): Production Security Engineering
Unit 8 (Weeks 13-16): Capstone Projects

Weekly Schedule

Unit 5: Multi-Agent Orchestration for Security (Weeks 1-4)

Week	Topic
Week 1: Multi-Agent Architecture Patterns	Single-agent limitations and multi-agent patterns
Week 2: CrewAI for Security Operations	Role-based multi-agent framework
Week 3: LangGraph for Stateful Workflows	State machines and incident response
Week 4: Agent Evaluation and Benchmarking	Quantitative metrics and testing

→ View Full Unit 5 Content

Unit 6: AI Attacker vs. AI Defender (Weeks 5-8)

Week	Topic
Week 5: Adversarial AI Threat Landscape	Threat modeling with MITRE ATLAS
Week 6: Red Teaming AI Agents	Offensive security assessment techniques
Week 7: Defending AI Agents	Guardrails, hardening, and defense strategies
Week 8: AI Attacker vs. Defender Wargame	Full security competition

→ View Full Unit 6 Content

Unit 7: Production Security Engineering (Weeks 9-12)

Week	Topic
Week 9: AI Supply Chain Security	Model provenance and dependency management
Week 10: Non-Human Identity Governance	Authentication and authorization for agents
Week 11: Observability and Cost Management	Production monitoring and optimization
Week 12: Deploying Agentic Systems	CI/CD, containerization, and operations

→ View Full Unit 7 Content

Unit 8: Capstone Projects (Weeks 13-16)

Week	Topic
Week 13: Capstone Kickoff and Architecture Reviews	Proposal and architectural design
Week 14: Capstone Development Sprint I	Building the core system
Week 15: Capstone Sprint II and Red Team Review	Hardening and peer security assessment
Week 16: Capstone Presentations	Final demo and reflection

→ View Full Unit 8 Content

Detailed Week Content

For detailed content including lecture notes, labs, and deliverables for each week, please visit the unit documentation files linked above. The remaining content below focuses on assessment, policies, and course resources.

Assessment Breakdown

Final grade calculation:

Component	Weight	Due Week(s)	Description
Lab Exercises (12 labs)	20%	1-12	Weekly lab deliverables and reports
Framework Comparison Report	10%	4	Unit 5 evaluation comparing Claude Agent SDK, CrewAI, LangGraph
Red Team Exercise	8%	6	Offensive security assessment of peer system
Blue Team Exercise	7%	7	Defensive hardening and response
Capstone Project	40%	13-16	Code, documentation, presentation, reflection
Peer Red Team Reviews	10%	15	Quality of security review work on peer projects
Metrics and Improvement Tracking	5%	13-16	Performance metrics throughout development

Letter Grade Scale:

A: 90-100%
B: 80-89%
C: 70-79%
D: 60-69%
F: Below 60%

Course Policies

Attendance and Participation

Attendance at all lectures is expected
Active participation in labs and team discussions is required
Excused absences (illness, family emergency, etc.) should be communicated to instructor as soon as possible
Unexcused absences may impact participation grade

Assignment Submission

All assignments due by end of stated deadline
Late submissions may be accepted with 10% deduction per 24-hour period, up to 3 days late
Assignments more than 3 days late receive zero credit (exceptions require instructor approval)
All code must be committed to course repository with clear commit messages

Academic Integrity

All work must be original
Collaboration is encouraged for labs, but individual understanding and contribution is required
Code/documentation must not be copied from external sources without attribution
Use of large language models (including Claude) is permitted and encouraged for coding assistance, but with proper attribution
Violations of academic integrity will result in failing the course and possible disciplinary action

Responsible Disclosure

Any vulnerabilities discovered in course assignments or capstone projects must be documented and disclosed only within the course environment
Findings may not be disclosed publicly or to external parties without instructor approval
Focus is on building security skills, not finding zero-days

Ethics of Offensive Security

All offensive security techniques (red teaming, exploitation, etc.) are practiced ONLY against course-provided systems or student-built systems
Absolutely no unauthorized testing against production systems or third-party systems
Students who violate this policy will fail the course and face disciplinary action
Focus is on defensive security principles and authorized assessments

Classroom Conduct

Respect diverse perspectives and backgrounds
Constructive criticism of ideas is encouraged; personal attacks are not
Maintain professional demeanor in all interactions
Violations of this policy will be addressed according to university conduct standards

Support and Accommodations

Students with documented disabilities should contact disability services and inform the instructor
Reasonable accommodations will be made to support equitable access to course materials
Additional tutoring and office hours are available for students needing support

Technology and Tools

All students should have access to:
A computer capable of running Docker, Python, and development tools
Anthropic Claude API access (Claude Max subscription preferred for full agentic capabilities)
GitHub account for code repositories
Cloud computing credits (AWS, GCP, or Azure) for deployment (if needed)
Cost of tools is student responsibility; financial hardship should be discussed with instructor

Course Resources and Readings

Required Texts and Documentation

Claude Agent SDK Documentation: https://docs.anthropic.com/agents
Model Context Protocol (MCP) Specification: https://modelcontextprotocol.io/
CrewAI Documentation: https://docs.crewai.com/
LangGraph Documentation: https://langchain-ai.github.io/langgraph/
MITRE ATLAS Framework: https://atlas.mitre.org/

Tools and Frameworks

Claude Code (IDE and code execution)
Claude Agent SDK (building agentic systems)
CrewAI (role-based multi-agent framework)
LangGraph (stateful workflow orchestration)
OpenTelemetry (observability)
Docker and containerization tools
CI/CD platforms (GitHub Actions, GitLab CI, etc.)
Vulnerability scanning tools — open-source first: Safety, pip-audit, OWASP Dependency-Check, Trivy; enterprise option: Snyk
Monitoring and logging platforms
Promptfoo and custom evaluation harnesses

Professional Organizations and Resources

SANS Institute: cybersecurity training and resources
(ISC)²: CISSP and other security certifications
Cybersecurity and Infrastructure Security Agency (CISA): threat intelligence
OpenSSF (Open Source Security Foundation): secure development practices
Anthropic and other AI safety/security research organizations

Contact and Support

Instructor Office Hours:

Times and location TBD
Drop-in availability or by appointment
Available for technical questions, capstone guidance, career advice

Teaching Assistants:

Available for lab support and technical questions
Hours TBD

Course Communication:

Primary: Course management system (Slack, Teams, or similar)
Email: For formal communications and grade disputes
GitHub Issues: For technical questions and lab setup issues

Mental Health and Wellness:

University counseling services: [contact information]
Peer support groups: [contact information]
If you're struggling, reach out to instructor or advisors

Course Modifications

This syllabus represents the instructor's current plans and intentions for this course. The instructor reserves the right to make reasonable adjustments to topics, readings, assessments, and policies during the semester to enhance student learning, address emergent topics in agentic AI security, or accommodate unforeseen circumstances. Students will be notified of any significant changes.

Course Version: 1.0 Last Updated: March 4, 2026 Next Review: September 2026