Internals#

How Claude Code works under the hood – the system prompt, context window, caching, token economics, and reasoning.

  • The System Prompt – What Claude reads before you say anything. Anatomy of the system prompt, how it’s assembled from multiple sources, and why it’s the largest factor in per-message token overhead.
  • Context Management – Working within the token budget. How the context window fills with system prompt, conversation history, tool results, and file contents, and strategies for managing it as sessions grow.
  • Prompt Caching – Why your system prompt doesn’t cost what you think. How the API reuses previously processed prompt prefixes to reduce both cost and latency.
  • Token Optimization – Auditing and optimizing the baseline token cost from skills, plugins, and system configuration that every session carries.
  • Extended Thinking – How Claude reasons through complex problems. Adaptive thinking, effort levels, thinking token billing, and cost management.
  • Tool Execution Context – Why skill instructions and CLAUDE.md rules may not be in active scope when Claude processes bash tool results, and how to structure workflows that depend on behavioral constraints.
The System Prompt Forward