// Track

Context Engineering & CLAUDE.md Architecture

Build the Layer That Actually Runs the Model

Context is the dominant lever — more impact than model selection, prompt wording, or tools combined. Master CLAUDE.md architecture, progressive disclosure, semantic memory with semantic memory layer, context rot detection, and the full playbook for a system that compounds with every session.

Recommended: Complete Advanced Claude Code first

11 lessons~86 min total

Lessons are shown in recommended order. Complete them in sequence for the best experience — or jump to any lesson.

Lesson 368·7 min read

Why Context Is the Real Product

The model is a commodity. Your CLAUDE.md, memory files, and Semantic Memory Layer are the competitive moat — context is what you ship, not just what you configure.

Lesson 369·8 min read

The Context Window as a Budget

Every token in a context window is spent, not stored — and conversation history reloads completely on every turn, making the reread tax your largest hidden cost.

Lesson 370·8 min read

CLAUDE.md Architecture

Four levels of context inheritance — enterprise, project, sub-directory, and session — each narrowing scope and overriding the outer layer on conflict.

Lesson 371·7 min read

Tiered Importance — What Belongs Where

Not everything belongs in CLAUDE.md — the 200-line discipline exists because Tier 1 rules get buried when Tier 2 and Tier 3 content floods the file.

Lesson 372·8 min read

Progressive Disclosure — Load-on-Demand Context

Load context at the last responsible moment — skills, deferred tools, and on-demand sub-docs that load only when the task actually needs them.

Lesson 373·8 min read

Memory Files & Persistent State

MEMORY.md is the auto-memory index. lessons.md captures what went wrong and the rule to prevent recurrence. Together they make every session smarter than the last.

Lesson 374·9 min read

Building a Knowledge OS — The Semantic Memory Layer Pattern

Semantic memory across 43 repos and 2,968 indexed chunks — the query-first habit, embeddings vs structured recall, and how to build a knowledge OS that improves with every session.

Lesson 375·8 min read

Context Rot & Pollution

Stale context doesn't just fail to help — it actively misleads. How to detect date mismatches, contradictions, orphaned references, and deprecated tool mentions before they cause damage.

Lesson 376·7 min read

Pruning & Compaction — When Forgetting Is a Feature

Pruning is not loss — it is precision. The four-question decision tree for every piece of context, the memory lifecycle from active to archived to forgotten, and contradiction detection.

Lesson 377·7 min read

Scoping Context to the Task

Domain scoping, sub-agent context isolation, and why a focused 80-line context for a trading agent beats a noisy 400-line context that includes design rules and content pipeline instructions.

Lesson 378·9 min read

The Context Engineering Playbook

The complete system: how all five context layers combine into a production-grade setup, and the hands-on audit checklist to verify your own CLAUDE.md and memory setup against the playbook standard.