⭐ If you find this repository helpful, please consider giving it a ⭐ here on GitHub (click the star button in the top right corner) It's a quick way to show support for this openly available code. ⭐

Harness Engineering

This is the corresponding code for the book Harness Engineering: Optimization, Coordination, and Adaptive Agentic Systems.

TOC

• Chapter 1 Why Harness Engineering Matters
• Chapter 2 The Harness as the Agent System Substrate
• Chapter 3 Skills, Instructions, and Runtime Scaffolding
• Chapter 4 Context-Scoped Skill Compilation
• Chapter 5 Agent Identity: The Durable Boundary
• Chapter 6 The Action Surface: Tools, Sandboxes, and Capability Scoping
• Chapter 7 Coordination as State: Why Multi-Agents Systems Fail at the Handoff Layer
• Chapter 8 Observability, Traces, and Harness Telemetry
• Chapter 9 Harness Optimization Patterns
• Chapter 10 Learning the Harness: RL over Skills, Routes, and Signals
• Chapter 11 Teaching the Harness to Evolve: Skill Internalization and Latent Coordination
• Chapter 12 Governance, Trust Boundaries, and the Harness in Production

Instructions and Navigation

All of the code is organized into folders. Each folder starts with CH followed by the chapter number. For example, CH01. The notebooks are then organized as follows: ch01_prompt_context_harness_progression.ipynb, where ch01 indicates the chapter and prompt_context_harness_progression what is done in the notebook.

Repo structure

├── LICENSE
├── README.md             <- The top-level README for developers using this project.
├── CH01                  <- Per chapter folder with Jupyter notebooks.
    ├── [name].ipynb      <- Jupyter notebooks with naming as mentioned above.
├── CH02                  <- Per chapter folder with Jupyter notebooks.
...                       <- Same structure for all chapters.
├── utils                 <- Custom classes and functions and utility functions.
├── resources             <- Some miscellaneous resources.

Running the Notebooks

Every notebook can be opened and run on Google Colab directly from the links below. Just click the Open In Colab badge next to the notebook you want to run.

Chapter 1 — Why Harness Engineering Matters

Notebook	Colab
Prompt & Context Harness Progression

Chapter 2 — The Harness as the Agent System Substrate

Notebook	Colab
Harness Components
State Across Execution Boundaries

NOTE: You may need to run the notebooks with a GPU.