The Context Window, Explained

Every conversation with Claude Code operates within a context window — Claude's short-term memory for the current session. Understanding how this works is one of the most important foundations for building effectively with an AI agent. In this lesson, you'll learn what the context window actually is, how tokens fill it up, and why it's not just your messages competing for space — system prompts, tools, file reads, and your claude.md file all take up room. You'll also learn how to monitor your usage with the /context command and by setting up a custom status line so you always know where you stand.

But managing your context window isn't just about fitting under a token limit — it's about keeping the signal clean. Too much noise from unrelated tasks can confuse Claude, and losing key details after a compaction can lead to forgotten instructions. This lesson covers practical strategies for staying ahead of both problems, including when to compact proactively (with custom instructions to preserve important details), when to clear entirely, and how auto-compact works as a safety net.

Topics covered:

• What the context window is and how tokens work
• Everything that fills your context window (messages, file reads, system prompt, tools, claude.md, and more)
• How claude.md always loads vs. Skills loading on demand
• Subagents and their separate context windows
• Monitoring context with /context and setting up /statusline
• The auto-compact buffer and how compaction works
• Why context quality matters — noise vs. missing information
• Using /compact with custom instructions, /clear, and auto-compact strategically