Runtime · 6 Chapters

Harness Engineering

The runtime layer beneath the model — the loop, the context window, compaction, and the system prompt. Not Claude Code features (that is the Toolkit). Not portable techniques (that is Patterns). The floor the other two stand on.

The Six Chapters

Read top to bottom — this is the floor, and each chapter assumes the one before it.

The Agent Loop

The core run loop: prompt, model, tool calls, results, repeat — and the stop conditions that end it.

AvailableRead

Context Window Economics

The token budget as the scarce resource: what occupies the window, the cost model, and what to evict.

AvailableRead

Compaction & Continuity

What survives a context-overflow boundary: summarization fidelity, the handoff, and designing for it.

AvailableRead

Tool Result Curation

The signal-to-noise problem: verbose tool output, truncation, deferral, and protecting the window.

AvailableRead

System Prompt Architecture

Where steering actually lives: instruction hierarchy, the harness preamble, and layered precedence.

AvailableRead

Building a Harness

Rolling your own loop with the Claude Agent SDK — when it pays off, and when to use the managed one.

AvailableRead

Where This Connects

Harness Engineering is the floor under the other two pillars. These patterns and toolkit deep-dives are the same ideas at the layer above.

Context PrimingPattern Memory LayerPattern Checkpoint LoopPattern Memory SystemToolkit Agents & SubagentsToolkit CLAUDE.mdToolkit