Agents · Runtimes and Harnesses

Explore the runtime core of an agent loop: define the harness boundary, then examine seven subsystems—state machines, context assembly, tool execution, permissions and sandboxing, persistence and recovery, human-in-the-loop handling, and observability.

Chapters

1. Chapter 16: Agent Harnesses—Definitions, Boundaries, and Layers
2. Chapter 17: The Agent Loop and Runtime State Machine
3. Chapter 18: Context Engineering and Context Assembly
4. Chapter 19: Tool Registries, Call Contracts, and Execution Pipelines
5. Chapter 20: Permissions, Sandboxing, and Isolation
6. Chapter 21: Checkpoints, Persistence, Retries, Timeouts, Idempotency, and Recovery
7. Chapter 22: Human-in-the-Loop Approval and Interruptible Execution
8. Chapter 23: Tracing, Evaluation, Cost Control, and Coding-Agent Harness Case Studies

Back to Agent Topics.