Skip to content

🌐 日本語

Part 2: Understanding Context Window

NOTE

Learn the structure of the LLM's "thinking space." The physical foundation for "why" the structural problems you learned in Part 1 occur lies here. All design decisions from Part 3 onward are based on the constraints of this context window.

What is Context Window?

An LLM thinks within a single "context window." For Claude, this is typically 200K to 1M tokens (depending on model generation). All configuration files in Claude Code are mechanisms to control "when, what, and how much" to inject into this window.

What You'll Learn in This Part

DocumentContent
Token・Context・Context WindowThree foundational concepts. Prerequisites for understanding this entire repository
Chat / SessionThe "temporal container" where context accumulates. Physical explanation for why context grows
What is Context WindowThe complete picture of what the LLM "sees"
Injection Timing OverviewWhen and how each configuration file enters the context
Context Budget as a StrategyHow to allocate finite tokens strategically

Why It's Between Part 1 and Part 3

In Part 1, you learn "LLMs have these problems." In Part 2, you understand "the LLM's thinking space has this structure." Then in Part 3 onward, you proceed to "therefore, configure it this way."

This follows the same order as understanding the Node.js runtime before writing TypeScript. Without knowing the runtime's constraints, you can't understand why code doesn't work.


Previous: Part 1: Understanding LLM Structural ConstraintsNext: Part 3: Always-Loaded Context

Released under the CC BY 4.0 License.