Anthropic 的 Claude Code 原始碼外洩後，多項值得關注的產品細節：

• Slash Commands 提升工作流程效率
/plan（先產生計劃）、/compact（壓縮對話）、/cost（即時查看成本）
• CLAUDE.md 作為核心行為規範
相當於「AI 工作守則」，透過明確規則確保 Claude 按標準流程運作
• 完善的安全與權限分級
清楚區分自動批准、需用戶確認、以及永不允許的工具操作
• 內部模型代號曝光
包括 Capybara、Fennec、Numbat 等未公開型號
• 多項仍未啟用的功能
語音模式、後台模式、遠端控制、自動任務排程等，顯示未來擴展空間相當大

🔍 Following the recent leak of Anthropic’s Claude Code source files, several product details have surfaced:
• Slash Commands that streamline workflows
/plan (generate a plan first), /compact (compress conversation context), /cost (monitor usage cost in real time)
• CLAUDE.md as the core behavioral contract
Effectively an “AI operating manual” that enforces strict rules and ensures Claude follows defined workflows
• A well‑structured security and permission model
Clear separation between auto‑approved actions, user‑confirmed actions, and permanently restricted operations
• Internal model codenames revealed
Including Capybara, Fennec, Numbat — models not previously disclosed publicly
• Multiple capabilities not yet enabled
Voice mode, background mode, remote execution, automated task scheduling — all indicating significant future expansion potential