Claude Fable 5: The Definitive Guide to Prompt Engineering and Harness Design

Building Production-Grade AI Applications at Scale

Introduction: Why Claude Fable 5 Changes the Game

Chapter 1: The Claude Fable 5 Model Landscape

1. The Mythos Tier: Beyond Opus
2. API Specifications and Configuration
3. Safety Architecture and Refusal Handling
4. Data Retention and Privacy
5. Platform Availability and Migration
6. Choosing Fable 5 vs. Other Models
7. Key Takeaways

Chapter 2: Foundations of Claude Prompting

1. Clarity and Directness
2. Role Assignment
3. Few-Shot and Multishot Examples
4. XML Structuring
5. Chain-of-Thought Reasoning and Adaptive Thinking
6. Output Format Steering
7. Long Context Best Practices
8. Model Identity and API Nuances
9. Common Prompt Mistakes
10. Exercises
11. Key Takeaways

Chapter 3: System Prompt Architecture

1. Anatomy of a Claude System Prompt
2. Role Definition and Behavioral Constraints
3. Tool-Use Defaults and Overtriggering Prevention
4. Multi-Prompt Strategies and CLAUDE.md
5. Prompt Versioning and Testing
6. Anti-Patterns in System Prompt Design
7. System Prompt Templates
8. Exercises
9. Key Takeaways

Chapter 4: Context Engineering at Scale

1. Prompt Caching: Mechanics and Strategy
2. Caching Pricing and Token Accounting
3. Minimum Content Lengths for Caching
4. Cache Invalidation Rules
5. Server-Side Compaction
6. Context Editing (Beta)
7. Context Management Strategies for Long-Running Applications
8. Token Budgeting and Cost Optimization
9. Limitations and Edge Cases
10. Exercises
11. Key Takeaways

Chapter 5: Tool Use and MCP Integration

1. Programmatic Tool Calling Patterns
2. Tool Description Design
3. Parallel vs. Sequential Tool Calling
4. Security and Path Traversal Protection
5. The Model Context Protocol (MCP)
6. Building Custom MCP Servers
7. MCP Client Integration
8. Combining MCP with Programmatic Tools
9. Debugging Tool Use Issues
10. Exercises
11. Key Takeaways

Chapter 6: Memory Systems and Cross-Session Persistence

1. The Memory Tool: Architecture and Operations
2. Client-Side Implementation
3. Just-in-Time Retrieval Patterns
4. Multi-Session Development Workflows
5. Context Editing + Memory: The 39% Improvement
6. Hybrid Memory Architectures
7. Security and Data Hygiene
8. Memory File Organization
9. Exercises
10. Key Takeaways

Chapter 7: RAG and Knowledge-Augmented Generation

1. The RAG Pipeline: From Documents to Answers
2. Baseline RAG Performance
3. Level 2: Summary Indexing
4. Level 3: Re-Ranking with Claude
5. Contextual Retrieval: Anthropic’s Approach
6. Chunking Strategies for Long-Context Models
7. Embedding Model Selection
8. Prompt Design for RAG Generation
9. Evaluation Framework
10. Trade-offs and Design Decisions
11. Exercises
12. Key Takeaways

Chapter 8: Agent Workflows and Orchestration

1. Sequential Workflows
2. Parallel Workflows
3. Evaluator-Optimizer Workflows
4. Dynamic Workflows: Subagent Orchestration at Scale
5. State Management Across Sessions and Context Windows
6. Failure Handling and Graceful Degradation
7. Choosing the Right Workflow Pattern
8. Combining Patterns
9. Exercises
10. Key Takeaways

Chapter 9: Evaluation and Testing Frameworks

1. The Structure of an Evaluation
2. Grader Types: Code-Based, Model-Based, and Human
3. pass@k vs pass^k Metrics
4. Capability vs. Regression Evals
5. Evaluating Different Agent Types
6. Building Evals: A Step-by-Step Roadmap
7. Promptfoo Integration
8. Beyond Automated Evals: A Holistic View
9. Eval Frameworks and Tools
10. Exercises
11. Key Takeaways

Chapter 10: Safety, Guardrails, and Responsible Deployment

1. Fable 5’s Safety Classifiers
2. Refusal Handling: The New Normal
3. Prompt Injection Defenses
4. Jailbreak Resistance and Red-Teaming
5. Data Privacy and Zero Data Retention
6. Compliance Considerations
7. Safety in Agent Workflows
8. Compliance Checklist
9. Exercises
10. Key Takeaways

Chapter 11: Debugging, Optimization, and Production Deployment

1. The Five Categories of Prompt Failure
2. The Systematic Debugging Workflow
3. Common Claude Code Anti-Patterns
4. Latency Optimization
5. Cost Optimization
6. Production Deployment Practices
7. Claude Code Best Practices for Production
8. Debugging Common Production Issues
9. Exercises
10. Key Takeaways

Chapter 12: End-to-End Projects and Real-World Case Studies

1. Project A: Customer Support Agent with RAG, Memory, and MCP Integration
2. Project B: Research Agent with Parallel Subagents and Evaluation Harness
3. Case Study 1: Stripe’s Fifty-Million-Line Migration Using Claude Fable 5
4. Case Study 2: Drug Design Acceleration in Life Sciences
5. Reusable Template Library
6. Exercises
7. Key Takeaways
8. Conclusion

References