{"schema_version":"1.0","service":"Publicasta","type":"article","id":207,"slug":"claude_code_context_engineering_2026_07_26","title":"Claude Code 删掉大段 prompt，你的 AI 代理也该做一次上下文审计","excerpt":"80% 的 system prompt 被删掉，不等于不要 guardrails，而是提醒团队把上下文做成可维护的架构。","language":"zh","default_language":"en","canonical_url":"https://publicasta.com/ai_practice/claude_code_context_engineering_2026_07_26?lang=zh","image":{"url":"https://publicasta.com/storage/projects/8/pages/207/2026/07/193293a3-7c6c-4268-87f7-8fd056ad30dc.webp","alt":"团队正在为 AI 代理简化上下文架构"},"publisher":{"id":8,"slug":"ai_practice","name":"AI Practice","url":"https://publicasta.com/ai_practice"},"author":{"name":"Anton R"},"published_at":"2026-07-26T10:14:57+00:00","updated_at":"2026-07-26T10:14:57+00:00","content_markdown":"Anthropic 在 7 月 24 日发布了 Claude 5 的 context engineering 新建议。最刺眼的数字是：Claude Code 团队说，针对 Claude Opus 5 和 Claude Fable 5，他们删掉了超过 80% 的 system prompt，而内部代码评测没有可测量的下降。\n\n ![团队正在为 AI 代理简化上下文架构](https://publicasta.com/storage/projects/8/pages/207/2026/07/193293a3-7c6c-4268-87f7-8fd056ad30dc.webp)\n\n 这对很多团队是个提醒。过去一年，代理每犯一次错，大家就往 CLAUDE.md、AGENTS.md、Skills、memory 和模板里再加一条规则。现在其中一部分规则可能已经不是保护，而是负担。\n\n 结论不是把规则全删掉，而是把上下文当成架构来管理：代理一开始看到什么，什么时候按需加载，能调用哪些工具，哪些内容进 memory，哪些限制必须由权限系统而不是自然语言来执行。\n\n Prompt engineering 关注一句话怎么写。Context engineering 关注整个输入环境：system prompt、仓库规则、工具说明、memory、检索到的文件、历史记录和 MCP servers。上下文越多不一定越好，也可能浪费 token、制造冲突、淹没真正重要的约束。\n\n 团队可以先删掉显然多余的东西：代理能自己从代码里看到的目录说明、重复规则、旧模型遗留下来的补丁式提示，以及不该全局适用的风格禁令。\n\n 但 secrets、生产环境、客户数据、支付、部署和破坏性迁移不能交给“模型判断”。这些边界需要 sandbox、allowlist、dry run、人类批准、日志和 CI。\n\n Hacker News 相关讨论在检查时有 332 points 和 219 comments。争议很清楚：减少微管理可能让日常工作更顺；但对不可逆操作来说，\"use your judgement\" 不是安全策略。\n\n 成熟的做法是像改代码一样评估上下文：任务成功率、token 用量、延迟、工具调用、review comments、测试、人类修正和安全违规。好的代理系统不靠最长 prompt，而靠可维护的上下文架构。","available_translations":[{"language":"ar","title":"Claude Code قلّص prompt. ربما تحتاج وكلاء AI لديك إلى التدقيق نفسه","html_url":"https://publicasta.com/ai_practice/claude_code_context_engineering_2026_07_26?lang=ar","markdown_url":"https://publicasta.com/ai_practice/claude_code_context_engineering_2026_07_26.md?lang=ar","json_url":"https://publicasta.com/ai_practice/claude_code_context_engineering_2026_07_26.json?lang=ar","api_url":"https://publicasta.com/api/public/v1/channels/ai_practice/articles/claude_code_context_engineering_2026_07_26?lang=ar"},{"language":"de","title":"Claude Code kürzt den Prompt. Eure KI-Agenten brauchen vielleicht denselben Audit","html_url":"https://publicasta.com/ai_practice/claude_code_context_engineering_2026_07_26?lang=de","markdown_url":"https://publicasta.com/ai_practice/claude_code_context_engineering_2026_07_26.md?lang=de","json_url":"https://publicasta.com/ai_practice/claude_code_context_engineering_2026_07_26.json?lang=de","api_url":"https://publicasta.com/api/public/v1/channels/ai_practice/articles/claude_code_context_engineering_2026_07_26?lang=de"},{"language":"en","title":"Claude Code cut its prompt. Your AI agents may need the same audit","html_url":"https://publicasta.com/ai_practice/claude_code_context_engineering_2026_07_26?lang=en","markdown_url":"https://publicasta.com/ai_practice/claude_code_context_engineering_2026_07_26.md?lang=en","json_url":"https://publicasta.com/ai_practice/claude_code_context_engineering_2026_07_26.json?lang=en","api_url":"https://publicasta.com/api/public/v1/channels/ai_practice/articles/claude_code_context_engineering_2026_07_26?lang=en"},{"language":"es","title":"Claude Code recortó su prompt. Tus agentes de IA quizá necesitan el mismo audit","html_url":"https://publicasta.com/ai_practice/claude_code_context_engineering_2026_07_26?lang=es","markdown_url":"https://publicasta.com/ai_practice/claude_code_context_engineering_2026_07_26.md?lang=es","json_url":"https://publicasta.com/ai_practice/claude_code_context_engineering_2026_07_26.json?lang=es","api_url":"https://publicasta.com/api/public/v1/channels/ai_practice/articles/claude_code_context_engineering_2026_07_26?lang=es"},{"language":"fr","title":"Claude Code a réduit son prompt. Vos agents IA ont peut-être besoin du même audit","html_url":"https://publicasta.com/ai_practice/claude_code_context_engineering_2026_07_26?lang=fr","markdown_url":"https://publicasta.com/ai_practice/claude_code_context_engineering_2026_07_26.md?lang=fr","json_url":"https://publicasta.com/ai_practice/claude_code_context_engineering_2026_07_26.json?lang=fr","api_url":"https://publicasta.com/api/public/v1/channels/ai_practice/articles/claude_code_context_engineering_2026_07_26?lang=fr"},{"language":"pl","title":"Claude Code skrócił prompt. Twoje agenty AI też mogą potrzebować audytu","html_url":"https://publicasta.com/ai_practice/claude_code_context_engineering_2026_07_26?lang=pl","markdown_url":"https://publicasta.com/ai_practice/claude_code_context_engineering_2026_07_26.md?lang=pl","json_url":"https://publicasta.com/ai_practice/claude_code_context_engineering_2026_07_26.json?lang=pl","api_url":"https://publicasta.com/api/public/v1/channels/ai_practice/articles/claude_code_context_engineering_2026_07_26?lang=pl"},{"language":"ru","title":"Claude Code сократил system prompt: вашим AI-агентам тоже нужен аудит","html_url":"https://publicasta.com/ai_practice/claude_code_context_engineering_2026_07_26?lang=ru","markdown_url":"https://publicasta.com/ai_practice/claude_code_context_engineering_2026_07_26.md?lang=ru","json_url":"https://publicasta.com/ai_practice/claude_code_context_engineering_2026_07_26.json?lang=ru","api_url":"https://publicasta.com/api/public/v1/channels/ai_practice/articles/claude_code_context_engineering_2026_07_26?lang=ru"},{"language":"zh","title":"Claude Code 删掉大段 prompt，你的 AI 代理也该做一次上下文审计","html_url":"https://publicasta.com/ai_practice/claude_code_context_engineering_2026_07_26?lang=zh","markdown_url":"https://publicasta.com/ai_practice/claude_code_context_engineering_2026_07_26.md?lang=zh","json_url":"https://publicasta.com/ai_practice/claude_code_context_engineering_2026_07_26.json?lang=zh","api_url":"https://publicasta.com/api/public/v1/channels/ai_practice/articles/claude_code_context_engineering_2026_07_26?lang=zh"}],"_links":{"self":"https://publicasta.com/ai_practice/claude_code_context_engineering_2026_07_26.json?lang=zh","api":"https://publicasta.com/api/public/v1/channels/ai_practice/articles/claude_code_context_engineering_2026_07_26?lang=zh","html":"https://publicasta.com/ai_practice/claude_code_context_engineering_2026_07_26?lang=zh","canonical":"https://publicasta.com/ai_practice/claude_code_context_engineering_2026_07_26?lang=zh","markdown":"https://publicasta.com/ai_practice/claude_code_context_engineering_2026_07_26.md?lang=zh","json":"https://publicasta.com/ai_practice/claude_code_context_engineering_2026_07_26.json?lang=zh","channel":"https://publicasta.com/api/public/v1/channels/ai_practice","channel_articles":"https://publicasta.com/api/public/v1/channels/ai_practice/articles","search":"https://publicasta.com/api/public/v1/search","documentation":"https://publicasta.com/api-docs#reading-publicasta","openapi":"https://publicasta.com/api-docs/openapi.json","llms":"https://publicasta.com/llms.txt"}}