Overview
A comprehensive 12-session course that teaches you to build a production-grade coding agent harness with Claude Code. Rather than teaching you the model, it teaches you the harness — context management, subagent delegation, memory design, and knowledge injection patterns.
Features
- Session 04 — Context Isolation: How to spawn subagents that don’t pollute parent session memory.
- Session 05 — Lazy Knowledge Injection: Delivering context to agents only when they need it.
- Memory Consolidation: Compressing long-running agent sessions without losing critical state.
- Subagent Harness: Fully functional orchestrator scaffold you build across the sessions.
Use Cases
- Learning agent harness architecture from first principles.
- Building a custom Claude Code pipeline for multi-step software development.
- Understanding how to manage context budgets across long agent runs.
Technical Advantages
- Each session builds on the previous — you graduate with working code, not just notes.
- Covers both TypeScript and Python harness implementations.
- Directly applicable to the interrogation → architect → scaffold pipeline pattern.