 

Claude Code System prompt占用过高问题解析与解决思路

2025-12-20 分类：前沿哨所阅读(2) 评论(0) 赞(0)

智谱 GLM，支持多语言、多任务推理。从写作到代码生成，从搜索到知识问答，AI 生产力的中国解法。

Claude Code用户在使用过程中发现System prompt占用空间超过20K，导致API使用量快速消耗。用户对比了不同环境下的表现：在WSL中(v2.0.54)打开时占用较高，而在Windows原生应用同目录(v2.0.74)打开时仅占用3K不到。用户询问System prompt具体包含什么内容，以及造成这种差异的原因。这一问题反映了AI工具在实际使用中的优化需求，对其他Claude Code用户具有参考价值。目前官方文档中未找到相关说明，需要社区进一步探讨解决方案。

原文链接：Linux.do

赞(0)

未经允许不得转载：Toy Tech Blog » Claude Code System prompt占用过高问题解析与解决思路

分享到

评论抢沙发

快讯

Hakimi Platform Delivers Effective Speaking Scoring: Gemini Models Show Strong Results

A user on the Linux.do forum shared their personal experience using the Hakimi platform to implement a speaking scoring model. After testing, Google's Gemini-3-flash-preview and Gemini-3-pro-preview models demonstrated strong accuracy, providing an effective solution for speaking assessment. This discovery highlights the potential of AI technology in language education, offering practical value for both language learners and educators. Three participants in the forum discussed the topic, further validating the feasibility of this method.

Original Link:Linux.do

29分钟前
Guide for Mainland China Companies Purchasing Cursor Team Edition

This article provides a detailed exploration of payment and invoicing solutions for mainland Chinese companies purchasing the Cursor Team Edition. It examines the feasibility of using Stripe payments, including Alipay and UnionPay card options, while addressing risk considerations for corporate account cards. The guide also resolves invoice issuance challenges for overseas procurement, offering practical solutions for tech companies to overcome AI tool purchasing obstacles and enhance team collaboration efficiency.

Original Link:Linux.do

29分钟前
Can AI Agents Handle Financial Empirical Work? A Technical Exploration

This article explores the feasibility of applying AI Agents in financial data cleaning and empirical research. As a finance professional with limited programming expertise, the author raises a practical question: can AI Agents be relied upon to complete financial data cleaning and empirical analysis? The article reflects the current hot topic of integrating AI technology with specialized fields, demonstrating the potential and challenges of AI in practical application scenarios. This discussion holds reference value for both technical professionals focusing on AI implementation and finance practitioners, warranting further exploration and research.

Original Link:Linux.do

29分钟前
Openrouter Officially Launches Claude Code Support

Recently, leading AI platform Openrouter announced the official launch of Claude code support, a development that has garnered significant attention from AI developers. Claude, the large language model developed by Anthropic, is renowned in the developer community for its exceptional coding capabilities. Openrouter's latest feature expansion provides developers with more diverse AI coding assistance options. After extensive preparation, this functionality is now available to users. Developers can now experience Claude's powerful code generation and debugging capabilities through the Openrouter platform, which will further enhance development efficiency and code quality.

Original Link:Linux.do

30分钟前
Altman: Google Remains Biggest Threat, OpenAI to Normalize Red Alerts

OpenAI CEO Sam Altman recently stated in an interview that despite massive investments and first-mover advantages, Google remains its "massive threat." Altman emphasized that maintaining "healthy paranoia" is essential for tech companies' survival and predicted that OpenAI will issue one to two "red alerts" annually, normalizing this crisis response mechanism. Although OpenAI has received billions in support from tech giants like NVIDIA and official institutions including the UK government, Altman believes Google's deep technical foundation and resource reserves make it the number one competitor. Additionally, he revealed that the strong emergence of Chinese AI company DeepSeek in early 2025 directly triggered an internal "red alert" at OpenAI, demonstrating that emerging forces can also heighten OpenAI's vigilance.

Original Link:Linux.do

30分钟前
Meta Accelerates AI Strategy: 'Mango' Model Targets Image and Video, Set for 2026 Release

According to The Wall Street Journal, Meta is developing a next-generation AI image and video model codenamed 'Mango', expected to launch in the first half of 2026. Meanwhile, Meta is also developing 'Avocado', a next-generation text-based large language model with a focus on enhancing programming capabilities. Meta's Chief AI Officer Alexandr Wang revealed in an internal meeting that the company is in the early stages of exploring 'world models', where AI learns environments by processing visual information. This summer, Meta reorganized its AI teams, establishing a Superintelligence Laboratory led by Wang and recruiting over 50 AI experts. These moves indicate Meta is accelerating its AI strategy to address increasingly intense tech competition, particularly in cutting-edge areas like multimodal AI and world models.

Original link:Linux.do

30分钟前