[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-1074":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":9,"language":9,"languages":9,"totalLinesOfCode":9,"stars":10,"forks":11,"watchers":12,"openIssues":13,"contributorsCount":14,"subscribersCount":14,"size":14,"stars1d":15,"stars7d":16,"stars30d":17,"stars90d":14,"forks30d":14,"starsTrendScore":16,"compositeScore":18,"rankGlobal":9,"rankLanguage":9,"license":9,"archived":19,"fork":19,"defaultBranch":20,"hasWiki":21,"hasPages":19,"topics":22,"createdAt":9,"pushedAt":9,"updatedAt":23,"readmeContent":24,"aiSummary":25,"trendingCount":14,"starSnapshotCount":14,"syncStatus":13,"lastSyncTime":26,"discoverSource":27},1074,"Hermes-Wiki","cclank\u002FHermes-Wiki","cclank","Hermes agent + LLM Wiki + 源代码 完成 Hermes agent  wiki ",null,539,66,3,2,0,13,39,155,9.48,false,"master",true,[],"2026-06-12 02:00:22","# Hermes Agent Architecture Wiki\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FWiki-Hermes_Agent-blue?style=for-the-badge&logo=markdown\" alt=\"Wiki\" height=\"28\">\n  \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FSource-hermes--agent-green?style=for-the-badge&logo=github\" alt=\"Source\" height=\"28\">\n  \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FKnowledge_Base-37_pages-orange?style=for-the-badge&logo=obsidian\" alt=\"Knowledge Base\" height=\"28\">\n  \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FVersion-v2026.4.23-purple?style=for-the-badge\" alt=\"Version\" height=\"28\">\n  \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FVerified-Source_Code-brightgreen?style=for-the-badge\" alt=\"Verified\" height=\"28\">\n  \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-MIT-yellow?style=for-the-badge\" alt=\"License\" height=\"28\">\n\u003C\u002Fp>\n\n> 基于 Nous Research [Hermes Agent](https:\u002F\u002Fgithub.com\u002FNousResearch\u002Fhermes-agent) 源码的深度架构文档。\n> 所有页面均经过**逐行源码验证**，确保准确性与时效性。\n\n\n---\n\n## 目录结构\n### 核心架构\n\n- [agent-loop-and-prompt-assembly](concepts\u002Fagent-loop-and-prompt-assembly.md): Agent 循环、系统提示构建、平台提示、执行指导\n- [tool-registry-architecture](concepts\u002Ftool-registry-architecture.md): 中央工具注册系统，声明式注册+集中调度\n- [model-tools-dispatch](concepts\u002Fmodel-tools-dispatch.md): 工具编排与调度，异步桥接+动态 schema 调整+参数类型强制\n- [toolsets-system](concepts\u002Ftoolsets-system.md): 工具分组系统、递归解析、14+ 平台工具集\n- [prompt-builder-architecture](concepts\u002Fprompt-builder-architecture.md): 系统提示模块化组装，注入防护+技能缓存+模型特定指导\n- [auxiliary-client-architecture](concepts\u002Fauxiliary-client-architecture.md): 辅助 LLM 客户端路由器，多 provider 解析链+自动降级\n- [provider-transport-architecture](concepts\u002Fprovider-transport-architecture.md): Provider Transport ABC，统一抽象 Anthropic\u002FChat Completions\u002FResponses API\u002FBedrock 的数据路径\n\n### 记忆与会话\n\n- [memory-system-architecture](concepts\u002Fmemory-system-architecture.md): 三层架构（MemoryStore\u002FMemoryManager\u002FMemoryProvider），冻结快照模式\n- [session-search-and-sessiondb](concepts\u002Fsession-search-and-sessiondb.md): FTS5 搜索 + LLM 摘要的跨会话回忆，orphan 删除策略\n- [context-compressor-architecture](concepts\u002Fcontext-compressor-architecture.md): 自动上下文压缩 v3，三阶段预处理（MD5 去重\u002FSmart Collapse\u002F参数截断）+ 结构化摘要 + OpenClaw 对比\n- [skills-and-memory-interaction](concepts\u002Fskills-and-memory-interaction.md): Skills 与 Memory 的互补关系和决策树\n- [skills-system-architecture](concepts\u002Fskills-system-architecture.md): 渐进式披露架构，技能发现、条件激活、密钥管理、插件命名空间技能、Curator 后台维护\n\n### 工具与能力\n\n- [browser-tool-architecture](concepts\u002Fbrowser-tool-architecture.md): 多后端浏览器自动化，accessibility tree+三层安全防护\n- [web-tools-architecture](concepts\u002Fweb-tools-architecture.md): 多后端搜索\u002F提取\u002F爬取，LLM 智能内容压缩\n- [code-execution-sandbox](concepts\u002Fcode-execution-sandbox.md): execute_code 沙箱，7 工具限制+UDS\u002FFile RPC 两种通信模式\n- [voice-mode-architecture](concepts\u002Fvoice-mode-architecture.md): Push-to-talk 语音交互，STT（3 Provider）+ TTS（5 Provider，含 Gemini\u002FxAI TTS）\n- [context-references](concepts\u002Fcontext-references.md): @file\u002F@folder\u002F@diff\u002F@url\u002F@git 引用系统，安全沙箱+注入量限制\n- [fuzzy-matching-engine](concepts\u002Ffuzzy-matching-engine.md): 8 策略链模糊匹配，从精确到相似度匹配\n- [large-tool-result-handling](concepts\u002Flarge-tool-result-handling.md): 三层溢出防护（工具内截断\u002F单结果持久化\u002F轮次聚合预算）\n\n### 性能与优化\n\n- [parallel-tool-execution](concepts\u002Fparallel-tool-execution.md): 智能并发安全检测，三层分类+路径冲突检测\n- [prompt-caching-optimization](concepts\u002Fprompt-caching-optimization.md): 冻结快照保护 prefix cache，75% 成本节省\n- [smart-model-routing](concepts\u002Fsmart-model-routing.md): 智能模型路由，短消息走便宜模型，AWS Bedrock\u002FGemini OAuth\u002FOllama Cloud\u002FTool Gateway\n\n### 安全与可靠性\n\n- [security-defense-system](concepts\u002Fsecurity-defense-system.md): 多层防御体系 + 危险命令审批系统（manual\u002Fsmart\u002Foff 三模式）\n- [interrupt-and-fault-tolerance](concepts\u002Finterrupt-and-fault-tolerance.md): 中断传播、结构化错误分类（error_classifier）、Fallback 模型链\n- [credential-pool-and-isolation](concepts\u002Fcredential-pool-and-isolation.md): 多密钥自动轮换、4 种选池策略、Profile 隔离\n\n### 多 Agent\n\n- [multi-agent-architecture](concepts\u002Fmulti-agent-architecture.md): 4 种运行时机制（delegate_task\u002FMoA\u002FBackground Review\u002Fsend_message）\n- [configuration-and-profiles](concepts\u002Fconfiguration-and-profiles.md): 多 Profile 架构，完全隔离的 agent 实例（第二种多 Agent 方案）\n\n### 平台与扩展\n\n- [cli-architecture](concepts\u002Fcli-architecture.md): CLI 架构、斜杠命令、hermes dump\n- [terminal-backends](concepts\u002Fterminal-backends.md): 7 种终端后端（含 Vercel Sandbox）、统一 spawn-per-call 执行模型\n- [messaging-gateway-architecture](concepts\u002Fmessaging-gateway-architecture.md): 18+ 平台统一网关（含 IRC\u002F腾讯元宝\u002F微信\u002FQQ Bot），平台适配器插件化（PlatformRegistry）、代理模式、channel_prompts、角色权限\n- [gateway-session-management](concepts\u002Fgateway-session-management.md): 网关会话管理，多平台会话隔离+PII 脱敏+重置策略\n- [hook-system-architecture](concepts\u002Fhook-system-architecture.md): 双 Hook 系统（Gateway Hooks + Plugin System），register_command\u002Fdispatch_tool，Dashboard 插件\n- [mcp-and-plugins](concepts\u002Fmcp-and-plugins.md): MCP 集成、插件钩子系统、OAuth 支持\n- [skin-engine](concepts\u002Fskin-engine.md): YAML 驱动的皮肤\u002F主题系统\n- [worktree-isolation](concepts\u002Fworktree-isolation.md): Git Worktree 并行隔离模式\n- [cron-scheduling](concepts\u002Fcron-scheduling.md): 内置调度器、自然语言调度、多平台投递\n- [trajectory-and-data-generation](concepts\u002Ftrajectory-and-data-generation.md): 轨迹保存、批量运行器、RL 训练环境\n\n### 更新日志\n\n- [2026-04-09-update](changelog\u002F2026-04-09-update.md): 59 commits，结构化错误分类、统一执行层、三层溢出防护、BlueBubbles 等\n- [2026-04-10-update](changelog\u002F2026-04-10-update.md): 293 commits，Context Engine 插件化、watch_patterns、WeChat、xAI、Discord\u002FSlack 增强\n- [2026-04-17-update](changelog\u002F2026-04-17-update.md): 641 commits (v0.10.0)，压缩 v3、Bedrock\u002FGemini\u002FOllama 新 Provider、Tool Gateway、插件命名空间技能、钉钉 QR 认证、Dashboard 插件\n- [2026-04-18-update](changelog\u002F2026-04-18-update.md): 410 commits post-v0.10.0，Transport ABC 重构、Shell Hooks、Delegate Orchestrator、Step Plan\u002FAI Gateway\u002FxAI STT\u002FKittenTTS、WeCom QR、Subagent 观测性\n- [2026-04-29-update](changelog\u002F2026-04-29-update.md): 182 commits (v2026.4.23)，平台适配器插件化（PlatformRegistry + IRC 参考实现）、Curator 后台技能维护、MiniMax OAuth、Vercel Sandbox、腾讯元宝、`on_session_switch`、`\u002Freload-skills`\n\n---\n\n## 统计信息\n\n- **概念页面**: 37 个\n- **更新日志**: 5 个\n- **源码覆盖**: 关键模块逐行验证\n- **跟踪版本**: v2026.4.23\n- **最后更新**: 2026-04-29\n\n\n## 使用方式\n\n- **GitHub 在线浏览**: 直接点击上方目录链接\n- **Obsidian 本地知识库**: \n  ```bash\n  git clone https:\u002F\u002Fgithub.com\u002Fcclank\u002FHermes-Wiki.git ~\u002FHermes-Wiki\n  ```\n- **配合 Hermes Agent**: 在 config.yaml 中设置 `skills.config.wiki.path: ~\u002FHermes-Wiki`\n\n\n---\n\n*本文档基于 Hermes Agent 源码分析生成。*\n","Hermes-Wiki 是一个基于 Hermes Agent 源码的深度架构文档项目，旨在为开发者提供详细的系统设计和实现指导。该项目通过逐行源码验证的方式确保了文档的准确性和时效性，涵盖了从核心架构到工具与能力等多个方面的内容。其技术特点包括但不限于多后端浏览器自动化、智能上下文压缩、安全沙箱执行代码等。适合于需要深入了解或使用 Hermes Agent 进行开发的研究人员和技术团队，以及对大型语言模型（LLM）相关技术感兴趣的开发者参考学习。","2026-06-11 02:41:29","CREATED_QUERY"]