Anthropic · CCAR-P
Context, tokens, and prompt reuse
CCAR-P · Lesson 6 of 18 · 40:04
YouTube loads only after you press play. This chapter opens at its timestamp in the full course. Watch on YouTube ↗
In this lesson
Prepare for the Claude Certified Architect – Professional exam, on the exam guide version 1.0, effective July 2026. In this lesson: the context window as a token budget, counting and trimming tokens, clearing old tool results and compacting, prompt caching (prefix order, automatic caching and breakpoints, cache lifetime and pricing, invalidation), modular prompts, and skills with progressive disclosure. Includes three original practice questions.
Preview the topics
- Harborline’s agent writes the current timestamp at the very start of its system prompt, on every request. Caching is enabled, but the hit rate is close to zero. Why?
- A long claims conversation keeps growing, with old tool results that are no longer needed, and the cost climbs with every turn. What helps most?
- Harborline has twelve detailed procedures, of a few thousand tokens each. With all of them in the system prompt, every request is long. Which approach fits?