Sync: add prd and ui system notes

wiki/overview.md

@@ -33,6 +33,8 @@ The wiki covers two major multi-agent frameworks: **The Agency** (agency-agents)
 
 **[[design-ux-architect]]**（ArchitectUX）：The Agency 设计部门的 UX 架构专家智能体——在 LuxuryDeveloper 实施前为其提供坚实的技术基础，包括 CSS 设计系统（变量令牌/排版层级/间距体系）、响应式布局框架（Grid/Flexbox/mobile-first）和 ThemeManager 三模式切换（light/dark/system）。核心原则：**Foundation-First**——在动手实现前先建立可扩展 CSS 架构，消除开发者架构决策疲劳；**默认强制要求**——所有新站点必须包含主题切换功能。ArchitectUX 作为 [[The Agency]] 设计部门智能体之一，承接 [[ProjectManager]] 任务列表并添加技术基础层，为 [[LuxuryDeveloper]] 输出清晰的交接规范。与 [[design-ui-designer]]（视觉细节）和 [[design-ux-researcher]]（用户研究）协同，遵循 [[Multi-Agent-System-Reliability]] 的交接协议模式。
 
+**[[design-whimsy-injector]]**（Whimsy Injector）：The Agency 设计部门的品牌个性化和愉悦感注入专家智能体——通过战略趣味设计为品牌体验增添个性、微交互和游戏化元素，使品牌通过意想不到的愉悦时刻脱颖而出。核心交付物：品牌个性框架（专业/休闲/错误/成功四种场景的人格光谱）、Whimsy 分类学（微妙/交互/发现/情境四类趣味）、微交互设计系统（CSS 动画 + 彩蛋 + 成就系统）。核心原则：**有目的的趣味**——每个趣味元素必须服务于功能或情感目的，不能分散用户注意力；**包容性愉悦设计**——确保趣味元素对残障用户可访问、不干扰屏幕阅读器、支持减少动画偏好。Whimsy Injector 与 [[ArchitectUX]] 互补——ArchitectUX 提供技术架构基础，Whimsy Injector 注入品牌人格，两者共同为 [[LuxuryDeveloper]] 提供完整的品牌体验设计。
+
 **[[multi-agent-system-reliability]]**（Alex Ewerlöf）：多智能体系统可靠性的架构模式理论——反对拟人化LLM，主张将LLM视为分布式系统中不可靠的组件。核心4模式：[[Hierarchy-Agent-Pattern]]（主管→工作→验证链）、[[Consensus-Voting-Pattern]]（N个LLM多数票消除幻觉）、[[Adversarial-Debate-Pattern]]（Generator→Critic→Judge对抗辩论）、[[Knock-out-Pattern]]（适者生存淘汰制）。核心洞察：不应要求模型"小心"，而应**强制**其正确——通过架构约束而非提示词约束。与 [[Designing for Agentic AI]] 互补（架构 vs 用户体验），与 [[Recursive Self-Optimization]] 共享自引用结构思想。与 [[Genetic-Algorithm]]（遗传算法）有关联——Knock-out/Tree of Thoughts 是 GA 的精简实现。
 
 Key concepts: [[Recursive Self-Optimization]], [[Generator Space]], [[Self-Referential Computation]], [[Fixed-Point Semantics]], [[Y-Combinator]], [[Self-Improving AI]], [[Automated Prompt Engineering]], [[Email Triage]], [[Newsletter Digest]], [[Preference Learning]], [[Cron Job]], [[Multi-Agent Coordination]], [[Multi-Tool Integration]], [[MCP Tool Interface Design]], [[Workflow Architecture]], [[Shared Memory Architecture]], [[Private Context]], [[Single Control Plane]], [[Scheduled Task Flywheel]], [[Parallel Agent Execution]], [[Topic-Based Routing]], [[Voice Interface]], [[Telephony Integration]], [[Voice Notification Channel]], [[Two-Way Voice Conversation]], [[Call-Worthy Threshold]], [[PSTN Calling]], [[PM Delegation Pattern]], [[CEO Pattern]], [[Shared State Coordination]], [[Git-as-Audit-Log]], [[Dynamic-Dashboard]], [[Alerting]], [[Zero-Friction Capture]], [[Cumulative Memory]], [[Conversational Interface]], [[Text-and-Search]], [[Unified-Inbox]], [[Intent-Classification]], [[Human-Handoff]], [[Test-Mode]], [[Business-Knowledge-Base]], [[Language-Detection]], [[AI-Auto-Response]], [[Heartbeat-Monitoring]], [[Self-Improving-Skill]], [[双层记忆架构]], [[每日复盘机制]], [[Pattern-Key]], [[Recurrence-Count]], [[Self-Improvement-Log]], [[AI-Agent思维方式]], [[批次任务拆分]], [[精确去重]], [[小文件清理]], [[安全删除策略]], [[Telegram通知]], [[Context-Window]], [[Model-Fallback]], [[Compaction]], [[Agent-Routing-Rules]], [[Error-Surface-vs-Root-Cause]], [[Layered-Configuration]], [[Log-Driven-Debugging]], [[Hidden-Failure-Paths]], [[Large Language Model]], [[RAG]], [[AI Agent]], [[ReAct Pattern]], [[Hierarchy-Agent-Pattern]], [[Consensus-Voting-Pattern]], [[Adversarial-Debate-Pattern]], [[Knock-out-Pattern]], [[Tree-of-Thoughts]], [[Genetic-Algorithm]], [[Reliability-Engineering]]