Anthropic published new guidelines on July 24 for context engineering with its Claude 5 generation models, focusing on how to optimize system prompts and other context sources. The update includes removing over 80% of the system prompt previously used in Claude Code to improve model performance, according to the company blog on claude.com.
The new rules emphasize that the prompt itself is only a small part of the overall context sent to Claude 5 models. Context is now assembled from multiple components including system prompts, Skills, CLAUDE.md files, and memory. This holistic approach to context engineering aims to enhance the quality and relevance of AI-generated outputs by better leveraging these combined inputs.
This development reflects Anthropic’s ongoing efforts to refine how users interact with its AI agents. By reducing reliance on lengthy system prompts and integrating diverse context elements, Claude 5 models can deliver more accurate and contextually appropriate responses. The approach aligns with broader trends in AI toward more dynamic and flexible prompt engineering techniques.
Anthropic’s blog post on July 24 details these changes and provides practical advice for developers working with Claude Code and Claude Enterprise. The company encourages users to apply these new context engineering principles to improve their AI applications.