Beyond prompt engineering: the full context window
A handbook and course for designing everything else the model sees—memory, retrieval, tools, and state.

What it does This repo is a structured handbook and emerging course for “context engineering”: the practice of curating the complete information payload fed to an LLM at inference time, not just the single prompt. It treats prompts as atoms and builds upward through few-shot examples, memory systems, multi-agent orchestration, and eventually field-theory abstractions. The material is organized as a progressive learning path with markdown foundations, Python walkthroughs, YAML templates, and working toy examples.
The interesting bit
The biological metaphor is deliberately ambitious—atoms to molecules to cells to organs to neural systems—yet the repo grounds it in concrete artifacts like minimal_context.yaml and a toy chatbot with context management. It also surfaces a survey-review of 1,400 research papers, giving the curriculum a claimed empirical backbone that most hand-wavy AI courses skip.
Key highlights
- Progressive curriculum: foundations → system implementation → integration → frontier topics (12 weeks outlined)
- Includes copy-paste templates (
20_templates/), runnable Python examples (10_guides_zero_to_hero/), and a toy chatbot implementation - Curated links to recent papers (ICML 2025, IBM Zurich, Princeton, MIT-Singapore) on memory architecture, tool integration, and emergent semantics
- Agent command files for Claude Code, OpenCode, Amp, Kiro, Codex, and Gemini CLI
- A mermaid diagram mapping prompt engineering (“what you say”) against context engineering (“everything else the model sees”)
Caveats
- The “Comprehensive Course” is explicitly marked “Under Construction”; much of the advanced material (quantum semantics, meta-recursive frameworks) appears aspirational
- The repo is primarily educational content and templates rather than a runnable framework or library
- Some of the dynamical-systems and field-theory framing reads more evocative than operational
Verdict Worth bookmarking if you’re building agent systems and want a principled vocabulary for memory, retrieval, and state management. Skip it if you need a drop-in SDK or mature tooling.