[{"data":1,"prerenderedAt":16},["ShallowReactive",2],{"category-blog-ai-agent":3,"category-docs-ai-agent":15},[4],{"path":5,"title":6,"description":7,"date":8,"category":9,"tags":10},"\u002Fblog\u002Fclaude-code-token-analysis","我分析了 99 个会话、76MB 日志，终于搞懂 Claude Code 的 token 到底烧在哪","用脚本量化本地会话日志后发现，输出只占成本约 12%，缓存重建才是 75% 的大头。记录三大烧钱黑洞、对应优化与可复用清单。","2026-07-17","ai-agent",[11,12,13,14],"claude-code","ai","performance","best-practice",[],1789635933370]