/ start-here
00 Start Here
The thesis. Why a model alone is not an agent, the precise difference between prompt, context, and harness engineering, and the five layers you will build across the workshop into your own coding harness.
→
What is a harness?
Claude Code is not a model. It is a harness. The layer of loop, tools, memory, and recovery wrapped around a model that turns transactional inference into a working agent.
→
Why 'just call the API' fails
A single completion is a transaction. An agent runs for two hundred turns, uses tools, recovers from errors, and remembers. The gap between them is the harness.
→
Prompt vs. context vs. harness engineering
Three disciplines, precise boundaries: what you say, what the model sees each turn, and the whole runtime the model lives inside. Context engineering is one subsystem of the harness.
→
The anatomy: five layers of a harnessMAP
The loop, tools + guardrails, the context engine, durability, and orchestration. The map for everything that follows, and for the harness you build.
→
What you will build: a pi-style harness
A terminal coding agent you own end to end, in the spirit of pi (pi.dev): it edits real files, asks before dangerous actions, compacts its own context, checkpoints every step, and dispatches sub-agents.
→
How to use this site
The two surfaces, the build-along code, and the order to read things in. How the book maps to the five live days.