Claude says it removed more than 80% of Claude Code’s system prompt for newer, more capable models, with no measurable performance loss. Their takeaway is that instructions that helped older models can become restrictive for stronger ones.
The six main changes were: give the model more room to use judgment, focus more on well-designed tools and interfaces, move optional instructions into skills through progressive disclosure, simplify tool descriptions and avoid repeating rules, use automatic memory, and provide richer references such as HTML artifacts, test suites, and code from other repositories.
The broader point: as models become more capable, better context and tool design may matter more than increasingly long, detailed prompts.
