Day 1 - Applications and Integration
Requirements & Systems Lifecycle
- Translating functional/business requirements into Claude‑powered application designs
- Systems lifecycle concepts: design → build → deploy → monitor
- Version control workflows (Git), code review practices
- Configuration management: CLAUDE.md, settings.json, plugin dependencies
Claude API Mechanics
- Messages API (roles, content blocks, tool calls)
- Streaming responses
- Vision inputs
- Thinking mode
- Caching API
- Batches API
- Third‑party invocation patterns
- Error handling and retry strategies
Software Engineering Foundations
- REST API fundamentals
- JSON schemas and structured output validation
- Async programming patterns (Python / TypeScript)
- Dependency management and environment setup
- Logging, observability, and monitoring Claude applications
Claude Application Design
- Designing multi‑interface Claude apps (claude.ai, Desktop, API, SDKs)
- Multi‑modal input handling (text, images, code)
- State management and context window budgeting
- Model version pinning and upgrade strategies
Day 2 - Model Selection and Optimization
LLM Fundamentals
- Tokens, tokenization, and context windows
- Sampling parameters (temperature, top‑p, top‑k)
- Determinism vs. non‑determinism
- Latency and throughput considerations
Model Tier Selection
- Capability tiers: Haiku, Sonnet, Opus, Fable (emerging)
- Choosing models based on task complexity, cost, and latency
- Structured output reliability across model tiers
Optimization Techniques
- Token‑efficient prompt design
- Caching strategies
- Batch processing for large workloads
- Cost management and monitoring
Agents and Workflows
Agent Architecture
- When to use agents vs. workflows
- Anthropic Agent SDK fundamentals
- Custom agent loops and harnesses
- Hosted vs. self‑hosted agent deployments
Agent Construction
- Building agents with Claude Agent SDK
- Memory management and context window strategies
- Tool‑use loops and multi‑step reasoning
- Sub‑agents and delegation patterns
Agent Patterns & Frameworks
- Strands
- LangGraph
- PydanticAI
- Common agentic design patterns (planner‑executor, router‑selector, tool‑first loops)
Day 3 - Prompt and Context Engineering
Prompt Engineering
- Instruction hierarchy and role prompting
- System vs. user vs. developer messages
- Multi‑turn prompt stability
- Guardrails through prompt structure
Context Engineering
- Context drift prevention
- Retrieval‑augmented context injection
- Chunking strategies for long documents
- Maintaining state across multi‑step workflows
Output Handling
- Enforcing structured output (JSON, XML, custom schemas)
- Validation and repair loops
- Using tool calls to constrain output
Tools and MCPs
Tool Implementation
- Tool schemas and function definitions
- Input/output validation
- Error handling inside tools
- Multi‑tool orchestration
MCP (Model Context Protocol)
- MCP server architecture
- Secure tool exposure
- Hooks and guardrails
- Defensive patterns against malicious or unsafe tool invocation
Custom Tools & Skills
- Custom tool development
- Integrating Skills with Claude Code
- Plugin dependencies and configuration
Security and Safety (8.1%)
Application Security
- Secrets and key management
- Secure API usage
- Environment isolation
- Logging and redaction of sensitive data
Prompt Injection Defense
- Injection patterns and detection
- Guardrail hooks
- Safe tool invocation
- Preventing destructive actions
Safety Principles
- Anthropic safety guidelines
- High‑risk request handling
- Model fallback behavior (e.g., classifier safeguard layers)
Claude Code
Core Components
- Rules
- Skills
- Commands
- Agents
- Agent Memory
Interaction Modes
- Session management
- Slash commands
- Headless mode
- Streaming mode
- Auto‑mode
Project Structure
- CLAUDE.md hierarchy
- Repository initialization
- settings.json configuration
- Plugin dependencies
Eval, Testing, and Debugging (2.6%)
Evaluation
- Designing evals
- Structured output validation
- Quality monitoring in production
- Regression testing for prompts and agents
Debugging
- Identifying error types (integration vs. model output)
- Trace analysis
- Failure isolation
- Recovery strategies
Testing
- Unit tests for tools and MCP servers
- Integration tests for agent workflows
- Load testing and stress scenarios
