{"id":"885f3794-f484-4426-bbfa-a85775332753","entityType":"agent","slug":"clawhub-wangm-a3-agent-cluster","name":"Agent Cluster","canonicalUrl":"https://www.xpersona.co/agent/clawhub-wangm-a3-agent-cluster","canonicalPath":"/agent/clawhub-wangm-a3-agent-cluster","generatedAt":"2026-10-11T20:56:57.468Z","source":"CLAWHUB","claimStatus":"UNCLAIMED","verificationTier":"NONE","summary":{"evidence":{"source":"editorial-content","verified":true,"confidence":"high","updatedAt":"2026-10-11T17:56:25.652Z","emptyReason":null},"description":"亚马逊外贸B2B多CMS Agent协作系统。支持Shopify/WooCommerce/Magento三大平台，A2A架构协调库存/采购/财务/物流四大专家Agent，三层安全网保障审批合规。即装即用，零配置开箱。 Skill: Agent Cluster Owner: wangm-a3 Summary: 亚马逊外贸B2B多CMS Agent协作系统。支持Shopify/WooCommerce/Magento三大平台，A2A架构协调库存/采购/财务/物流四大专家Agent，三层安全网保障审批合规。即装即用，零配置开箱。 Tags: latest:3.0.4 Version history: v3.0.4 | 2026-05-01T15:03:39.752Z | user Security fixes: removed eval/exec patterns, replaced exposed API key placeholders, cleaned prompt patterns v3.0.3 | 2026-04-29T03:50:57.891Z | user Remove hardcoded key patterns from test","descriptionLabel":"Technical summary","evidenceSummary":"Capability contract not published. No trust telemetry is available yet. 1K downloads reported by the source. Last updated 10/11/2026.","installCommand":"clawhub skill install s178tfk08y1fqgtvraxrr20g4h84g03e:agent-cluster","sourceUrl":"https://clawhub.ai/wangm-a3/agent-cluster","homepage":"https://clawhub.ai/wangm-a3/skills/agent-cluster","primaryLinks":[{"label":"View on ClawHub","url":"https://clawhub.ai/wangm-a3/agent-cluster","kind":"source"},{"label":"Homepage","url":"https://clawhub.ai/wangm-a3/skills/agent-cluster","kind":"homepage"}],"safetyScore":84,"overallRank":62,"popularityScore":60,"trustScore":null,"claimedByName":null,"isOwner":false,"seoDescription":"亚马逊外贸B2B多CMS Agent协作系统。支持Shopify/WooCommerce/Magento三大平台，A2A架构协调库存/采购/财务/物流四大专家Agent，三层安全网保障审批合规。即装即用，零配置开箱。 Skill: Agent Cluster Owner: wangm-a3 Summary: 亚马逊外贸"},"coverage":{"evidence":{"source":"public-profile","verified":false,"confidence":"medium","updatedAt":"2026-10-11T17:56:25.652Z","emptyReason":null},"protocols":[{"protocol":"OPENCLEW","label":"OpenClaw","status":"self-declared","notes":"Declared in the public agent profile."}],"capabilities":[],"verifiedCount":0,"selfDeclaredCount":1,"capabilityMatrix":{"rows":[{"key":"OPENCLEW","type":"protocol","support":"unknown","confidenceSource":"profile","notes":"Listed on profile"}],"flattenedTokens":"protocol:OPENCLEW|unknown|profile"}},"adoption":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"medium","updatedAt":"2026-10-11T17:56:25.652Z","emptyReason":null},"stars":null,"forks":null,"downloads":1014,"packageName":null,"latestVersion":"3.0.4","tractionLabel":"1K downloads"},"release":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"medium","updatedAt":"2026-10-11T17:56:25.637Z","emptyReason":null},"lastUpdatedAt":"2026-10-11T17:56:25.652Z","lastCrawledAt":"2026-10-11T17:56:25.637Z","lastIndexedAt":null,"nextCrawlAt":"2026-10-12T17:56:25.637Z","lastVerifiedAt":null,"highlights":[{"version":"3.0.4","createdAt":"2026-05-01T15:03:39.752Z","changelog":"Security fixes: removed eval/exec patterns, replaced exposed API key placeholders, cleaned prompt patterns","fileCount":145,"zipByteSize":522875},{"version":"3.0.3","createdAt":"2026-04-29T03:50:57.891Z","changelog":"Remove hardcoded key patterns from test files","fileCount":144,"zipByteSize":521505},{"version":"3.0.2","createdAt":"2026-04-29T03:37:14.835Z","changelog":"Add required_env declarations, fix placeholder keys","fileCount":144,"zipByteSize":521512},{"version":"3.0.1","createdAt":"2026-04-29T03:23:55.039Z","changelog":"Add required_env declaration for API keys to fix Suspicious flag","fileCount":144,"zipByteSize":521474},{"version":"1.0.0","createdAt":"2026-04-24T13:41:06.130Z","changelog":"foreign-trade-silicon-army v3.0.0 introduces a major update with enhanced multi-agent B2B collaboration, ERP integration, and robust error handling. - Added real ERP API integration layer supporting SAP, 用友, 金蝶, with circuit breaker, health checks, and mock mode. - Enhanced cross-agent B2B collaboration with fine-grained task protocol, state sync, and end-to-end traceability. - Integrated advanced error handling: 7-state task state machine, automatic exception classification, and flexible retry strategies. - Expanded to support Shopify, WooCommerce, and Magento for multi-CMS coordination. - Reinforced security and compliance with multi-layer approvals and audit logging. - Streamlined setup: plug and play, zero configuration required.","fileCount":144,"zipByteSize":521451}]},"execution":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No published capability contract is available yet."},"installCommand":"clawhub skill install s178tfk08y1fqgtvraxrr20g4h84g03e:agent-cluster","setupComplexity":"low","setupSteps":["Setup complexity is LOW. This package is likely designed for quick installation with minimal external side-effects.","Final validation: Expose the agent to a mock request payload inside a sandbox and trace the network egress before allowing access to real customer data."],"contract":{"contractStatus":"missing","authModes":[],"requires":[],"forbidden":[],"supportsMcp":false,"supportsA2a":false,"supportsStreaming":false,"inputSchemaRef":null,"outputSchemaRef":null,"dataRegion":null,"contractUpdatedAt":null,"sourceUpdatedAt":null,"freshnessSeconds":null},"invocationGuide":{"preferredApi":{"snapshotUrl":"https://www.xpersona.co/api/v1/agents/clawhub-wangm-a3-agent-cluster/snapshot","contractUrl":"https://www.xpersona.co/api/v1/agents/clawhub-wangm-a3-agent-cluster/contract","trustUrl":"https://www.xpersona.co/api/v1/agents/clawhub-wangm-a3-agent-cluster/trust"},"curlExamples":["curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-wangm-a3-agent-cluster/snapshot\"","curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-wangm-a3-agent-cluster/contract\"","curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-wangm-a3-agent-cluster/trust\""],"jsonRequestTemplate":{"query":"summarize this repo","constraints":{"maxLatencyMs":2000,"protocolPreference":["OPENCLEW"]}},"jsonResponseTemplate":{"ok":true,"result":{"summary":"...","confidence":0.9},"meta":{"source":"CLAWHUB","generatedAt":"2026-10-11T20:56:57.467Z"}},"retryPolicy":{"maxAttempts":3,"backoffMs":[500,1500,3500],"retryableConditions":["HTTP_429","HTTP_503","NETWORK_TIMEOUT"]}},"endpoints":{"dossierUrl":"https://www.xpersona.co/api/v1/agents/clawhub-wangm-a3-agent-cluster/dossier","snapshotUrl":"https://www.xpersona.co/api/v1/agents/clawhub-wangm-a3-agent-cluster/snapshot","contractUrl":"https://www.xpersona.co/api/v1/agents/clawhub-wangm-a3-agent-cluster/contract","trustUrl":"https://www.xpersona.co/api/v1/agents/clawhub-wangm-a3-agent-cluster/trust"}},"reliability":{"evidence":{"source":"runtime-metrics","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No trust, reliability, or runtime telemetry is available."},"trust":{"status":"unavailable","handshakeStatus":"UNKNOWN","verificationFreshnessHours":null,"reputationScore":null,"p95LatencyMs":null,"successRate30d":null,"fallbackRate":null,"attempts30d":null,"trustUpdatedAt":null,"trustConfidence":"unknown","sourceUpdatedAt":null,"freshnessSeconds":null},"decisionGuardrails":{"doNotUseIf":["Contract metadata is missing or unavailable for deterministic execution."],"safeUseWhen":[],"riskFlags":["missing_or_unavailable_contract","trust_data_unavailable","schema_references_missing"],"operationalConfidence":"low"},"executionMetrics":{"observedLatencyMsP50":null,"observedLatencyMsP95":null,"estimatedCostUsd":null,"uptime30d":null,"rateLimitRpm":null,"rateLimitBurst":null,"lastVerifiedAt":null,"verificationSource":null},"runtimeMetrics":{"successRate":null,"avgLatencyMs":null,"avgCostUsd":null,"hallucinationRate":null,"retryRate":null,"disputeRate":null,"p50Latency":null,"p95Latency":null,"lastUpdated":null}},"benchmarks":{"evidence":{"source":"no-benchmark-data","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No benchmark suites or observed failure patterns are available."},"suites":[],"failurePatterns":[]},"artifacts":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"high","updatedAt":"2026-10-11T17:56:25.652Z","emptyReason":null},"readme":"Skill: Agent Cluster\n\nOwner: wangm-a3\n\nSummary: 亚马逊外贸B2B多CMS Agent协作系统。支持Shopify/WooCommerce/Magento三大平台，A2A架构协调库存/采购/财务/物流四大专家Agent，三层安全网保障审批合规。即装即用，零配置开箱。\n\nTags: latest:3.0.4\n\nVersion history:\n\nv3.0.4 | 2026-05-01T15:03:39.752Z | user\n\nSecurity fixes: removed eval/exec patterns, replaced exposed API key placeholders, cleaned prompt patterns\n\nv3.0.3 | 2026-04-29T03:50:57.891Z | user\n\nRemove hardcoded key patterns from test files\n\nv3.0.2 | 2026-04-29T03:37:14.835Z | user\n\nAdd required_env declarations, fix placeholder keys\n\nv3.0.1 | 2026-04-29T03:23:55.039Z | user\n\nAdd required_env declaration for API keys to fix Suspicious flag\n\nv1.0.0 | 2026-04-24T13:41:06.130Z | auto\n\nforeign-trade-silicon-army v3.0.0 introduces a major update with enhanced multi-agent B2B collaboration, ERP integration, and robust error handling.\n\n- Added real ERP API integration layer supporting SAP, 用友, 金蝶, with circuit breaker, health checks, and mock mode.\n- Enhanced cross-agent B2B collaboration with fine-grained task protocol, state sync, and end-to-end traceability.\n- Integrated advanced error handling: 7-state task state machine, automatic exception classification, and flexible retry strategies.\n- Expanded to support Shopify, WooCommerce, and Magento for multi-CMS coordination.\n- Reinforced security and compliance with multi-layer approvals and audit logging.\n- Streamlined setup: plug and play, zero configuration required.\n\nArchive index:\n\nArchive v3.0.4: 145 files, 522875 bytes\n\nFiles: __init__.py (406b), _meta.json (132b), 30-expansion/30-agents-design.md (51910b), 30-expansion/agent_protocol.md (16120b), 30-expansion/agent_registry.yaml (18892b), 30-expansion/IMPLEMENTATION_PLAN.md (13710b), api_integration/__init__.py (936b), api_integration/api_adapter.py (21482b), api_integration/api_config.py (9243b), api_integration/api_health.py (8629b), api_integration/deepseek_integration.md (11899b), api_integration/mock_data.py (14178b), cms_approvals/04cf553c-e0c9-4732-870b-83f96e443b52.json (533b), cms_approvals/1c9e92a7-46f3-4a6d-a0d9-59c60ff0fee1.json (692b), cms_approvals/265648cd-fd95-4b2c-a1e7-4051cbe771cc.json (569b), cms_approvals/2c363c59-d353-4e3d-a2b6-aa4e321fa278.json (569b), cms_approvals/31ff6cde-b148-47d1-980f-6953a7e1fe98.json (569b), cms_approvals/326cf306-f324-478f-9abb-952657fd8a58.json (569b), cms_approvals/380f9c96-b301-4cbc-9a61-d508323241bf.json (717b), cms_approvals/5709cd14-a912-4bb9-a877-e8fd3c029ec2.json (569b), cms_approvals/a2941003-ecb3-41bc-8522-088b9bae1f41.json (701b), cms_approvals/b260ecbf-649d-4da4-b064-3093114d7ca7.json (704b), cms_approvals/d55ab274-327d-4090-985d-8c8c4d795c5c.json (683b), cms_approvals/de294f4f-a55b-43b2-bbae-3840b81f2e85.json (674b), cms_approvals/eb03a9c1-ecb6-4007-ada6-aaefd49ef4ac.json (533b), cms_executor/__init__.py (2412b), cms_executor/agent_integration.py (26443b), cms_executor/ARCHITECTURE.md (14538b), cms_executor/connectors/__init__.py (891b), cms_executor/connectors/amazon_connector.py (14762b), cms_executor/connectors/base_connector.py (27353b), cms_executor/connectors/magento_connector.py (16667b), cms_executor/connectors/shopify_connector.py (16183b), cms_executor/connectors/wordpress_connector.py (11068b), cms_executor/engine/__init__.py (642b), cms_executor/engine/approval.py (11726b), cms_executor/engine/audit.py (17904b), cms_executor/engine/executor.py (19384b), cms_executor/engine/rollback.py (15292b), cms_executor/tests/__init__.py (224b), cms_executor/tests/test_agent_integration.py (14358b), cms_executor/tests/test_approval.py (8408b), cms_executor/tests/test_base_connector.py (12002b), cms_executor/tests/test_connectors.py (14832b), cms_executor/tests/test_executor.py (13274b), cms_executor/tests/test_rollback.py (9323b), cms-executor/connectors/__init__.py (457b), cms-executor/connectors/amazon_connector.py (26421b), cms-executor/connectors/base_connector.py (5652b), cms-executor/connectors/shopify_connector.py (25896b), cms-executor/connectors/wordpress_connector.py (8616b), cms-executor/engine/__init__.py (168b), cms-executor/engine/approval.py (9193b), cms-executor/engine/rollback.py (10602b), cms-executor/README.md (12581b), cms-executor/tests/test_cms_executor.py (13197b), cms-executor/tests/test_connectors.py (31593b), collaboration/__init__.py (1067b), collaboration/state_sync.py (10976b), collaboration/task_protocol.py (15311b), collaboration/tests/__init__.py (22570b), collaboration/trace_tracker.py (11782b), collaboration/workflow_engine.py (12650b), config/agents.yaml (3021b), config/engines.yaml (7322b), config/permissions.yaml (3718b), config/workflows.yaml (3799b), error_handling/__init__.py (1260b), error_handling/exception_middleware.py (11348b), error_handling/operation_log.py (15910b), error_handling/retry_policy.py (10861b), error_handling/task_state_machine.py (13609b), execution/__init__.py (1564b), execution/circuit_breaker.py (40704b), execution/claude_ma_engine.py (8921b), execution/deepseek_engine.py (37869b), execution/engine_base.py (23172b), execution/engine_router.py (37840b), execution/gpt6_engine.py (27838b), execution/local_engine.py (11985b)\n\nFile v3.0.4:SKILL.md\n\n---\nname: foreign-trade-silicon-army\ndescription: 亚马逊外贸B2B多CMS Agent协作系统。支持Shopify/WooCommerce/Magento三大平台，A2A架构协调库存/采购/财务/物流四大专家Agent，三层安全网保障审批合规。即装即用，零配置开箱。\nversion: 3.0.4\ntriggers:\n  - \"外贸\"\n  - \"CMS\"\n  - \"Shopify\"\n  - \"WooCommerce\"\n  - \"Magento\"\n  - \"库存\"\n  - \"采购\"\n  - \"Amazon\"\n  - \"B2B\"\ntags:\n  - agent\n  - a2a\n  - cms\n  - ecommerce\n  - b2b\n  - foreign-trade\n  - multi-agent\nrequires:\n  python_packages:\n    - httpx\n    - pyyaml\n    - fastapi\n    - uvicorn\n  env:\n    - SYSTEM_MODE\n    - SAP_BASE_URL\n    - SAP_API_KEY\n    - YONYOU_BASE_URL\n    - YONYOU_APPKEY\n    - KINGDEE_BASE_URL\n    - ANTHROPIC_API_KEY\n    - DEEPSEEK_API_KEY\nhomepage: https://github.com/WangM-A3/agent-cluster\nauthors:\n  - WangM-A3\npricing: free,pro=9.9,enterprise\n---\n\n# 产业互联网硅基军团 v2.0\n\n> 企业级Multi-Agent智能体集群系统，基于**1+N架构**（1个幕僚长+20个专业Agent），参考OpenClaw Main Agent、腾讯ADP Router设计。v2.0全面升级：真实ERP API接入、协作流程细化、错误处理体系。\n\n## 核心能力\n\n### v2.0三大升级\n\n**1. 真实API接入层（api_integration/）**\n- 多ERP适配器：SAP S/4HANA、用友U8/NC/YonBIP、金蝶K3 Cloud/EAS、通用REST\n- 断路器模式（Circuit Breaker）+ 故障自动降级\n- 健康检查轮询（10s间隔，自动摘除异常节点）\n- 演示模式保留（MockDataGenerator，variance=0.1随机波动）\n\n**2. 跨Agent协作细化（collaboration/）**\n- 细粒度任务协议（TaskMessage）：依赖声明、优先级、TTL\n- 状态同步（SharedStateManager）：TTL+pub/sub通知机制\n- 全链路追踪（CollaborationTracker）：trace_id/span_id + Mermaid时序图可视化\n\n**3. 错误处理与状态管理（error_handling/）**\n- 7状态任务状态机：pending→running→success/failed/retry/timeout/cancelled\n- 10类异常自动分类：VALIDATION/NETWORK/TIMEOUT/AUTH/RESOURCE/NOT_FOUND等\n- 5种重试策略：FIXED/EXPONENTIAL/FIBONACCI/JITTER/ADAPTIVE\n- 敏感信息脱敏 + SOC2合规审计日志\n\n## 系统架构\n\n```\n用户请求 → Orchestrator（意图识别→任务拆解→智能体调度）\n    ↓\n20个专业Agent：采购/生产/销售/财务/运营/战略/研发/人力/合规\n    ↓\nAPI Integration Layer（v2.0新增）\n  ├─ SAP/用友/金蝶适配器（真实ERP）\n  └─ 断路器+健康检查+Mock降级\n```\n\n## 目录结构\n\n```\nagent-cluster/\n├── orchestrator.py              # 指挥智能体（核心调度器）\n├── api_integration/              # v2.0新增：真实API接入层\n│   ├── api_adapter.py           # 多ERP适配器（SAP/用友/金蝶）\n│   ├── api_config.py            # 配置化管理\n│   ├── api_health.py            # 健康检查+断路器\n│   └── mock_data.py             # 模拟数据（开发/演示）\n├── collaboration/                # v2.0新增：跨Agent协作\n│   ├── task_protocol.py         # 细粒度任务协议\n│   ├── state_sync.py            # 状态同步+TTL+pub/sub\n│   ├── trace_tracker.py         # 全链路追踪+Mermaid\n│   └── workflow_engine.py       # 混合执行引擎\n├── error_handling/              # v2.0新增：错误处理\n│   ├── task_state_machine.py    # 7状态任务状态机\n│   ├── exception_middleware.py  # 统一异常处理\n│   ├── retry_policy.py          # 5种重试策略\n│   └── operation_log.py         # 操作日志+脱敏+合规\n├── specialists/                 # 专业智能体\n│   ├── inventory_agent.py      # 库存智能体\n│   ├── logistics_agent.py       # 物流智能体\n│   ├── procurement_agent.py     # 采购智能体\n│   ├── finance_agent.py         # 财务智能体\n│   └── doc_agent.py            # 工艺文档智能体\n├── mcp_servers/                 # MCP协议封装\n│   ├── erp_server.py           # ERP系统接口\n│   ├── wms_server.py           # WMS仓库管理接口\n│   └── srm_server.py           # SRM供应商管理接口\n├── safety/                      # 安全围栏\n│   ├── permission_manager.py    # RBAC权限管理\n│   ├── audit_logger.py          # 全链路审计日志\n│   └── human_loop.py            # 人机回环审批\n└── config/                      # 配置文件\n    ├── agents.yaml             # 智能体定义\n    ├── workflows.yaml          # 工作流配置\n    └── permissions.yaml        # 权限矩阵\n```\n\n## 快速开始\n\n### 环境要求\n- Python 3.10+\n- 依赖：`pip install pyyaml fastapi uvicorn httpx aiofiles`\n\n### 运行\n```bash\ncd agent-cluster\npython orchestrator.py\n```\n\n### 配置（生产模式）\n```bash\nexport SYSTEM_MODE=production\nexport SAP_BASE_URL=https://sap.example.com\nexport SAP_API_KEY=sk-xxx\nexport YONYOU_BASE_URL=https://yonyou.example.com\n```\n\n## 触发词\n\n塑化报价 | 塑料原料采购 | 库存管理 | 生产排产 | 客户跟进 | 供应商比价 | B2B运营 | 工厂管理 | ERP集成 | 智能客服 | 行业KPI | 成本核算 | 硅基军团 | 工业Agent | 制造业AI | 产业互联网\n\n## 标签\n\n制造业AI, 产业互联网, Multi-Agent, ERP集成, 智能排产, 供应商管理, 报价系统, 智能工厂, AI运营, 企业数字化\n\n## 分类\n\n效率工具\n\n## 版本\n\nv2.0.0 - 三大升级：真实API接入层、协作流程细化、错误处理与状态管理\n\nFile v3.0.4:cms-executor/README.md\n\n# CMS Executor - 多平台 CMS 连接器\n\n## 概述\n\n本模块提供多平台（WordPress / Shopify / Amazon）CMS 直连执行能力，支持通过平台原生 API 创建/更新/删除内容，配合审批引擎和回滚机制实现安全的内容管理。\n\n## 目录结构\n\n```\ncms-executor/\n├── connectors/\n│   ├── __init__.py              # 统一导出\n│   ├── base_connector.py        # 抽象基类（所有 CMS 平台通用接口）\n│   ├── wordpress_connector.py   # WordPress REST API 实现\n│   ├── shopify_connector.py     # Shopify GraphQL Admin API 实现\n│   └── amazon_connector.py      # Amazon SP-API 实现\n├── engine/\n│   ├── __init__.py\n│   ├── approval.py              # 多级审批引擎\n│   └── rollback.py              # 变更追踪与回滚\n└── tests/\n    ├── test_cms_executor.py     # WordPress & 审批引擎测试\n    └── test_connectors.py       # Shopify & Amazon 连接器测试\n```\n\n---\n\n## 平台支持\n\n| 平台 | API | 认证方式 |\n|------|-----|---------|\n| WordPress | REST API v2 | Application Passwords |\n| Shopify | GraphQL Admin API 2024-01 | OAuth / API Key |\n| Amazon | SP-API | LWA OAuth |\n\n---\n\n## 1. Shopify 连接器\n\n### 认证配置\n\n1. 在 Shopify 后台创建私有应用或自定义应用\n2. 配置 API 权限（Products, Variants, Images）\n3. 获取访问令牌（Access Token）\n\n### 基本使用\n\n```python\nfrom connectors import ShopifyConnector, CMSCredential, ContentPayload\n\n# 初始化连接器\ncred = CMSCredential(\n    url=\"https://your-shop.myshopify.com\",\n    api_key=\"<YOUR_SHOPIFY_ACCESS_TOKEN>\",   # 访问令牌\n)\nshopify = ShopifyConnector(cred)\n\n# 验证连接\nif shopify.authenticate():\n    print(\"✅ Shopify 连接成功\")\n\n# ── 商品 CRUD ────────────────────────────────────────\nfrom connectors.shopify_connector import ProductPayload, VariantPayload\n\n# 方式一：使用 ContentPayload（兼容基类）\npayload = ContentPayload(title=\"My Product\", content=\"<p>Description</p>\", status=\"draft\")\nresult = shopify.create_content(payload)\nprint(f\"✅ 商品已创建，ID={result['id']}, handle={result['handle']}\")\n\n# 方式二：使用 ProductPayload（Shopify 原生字段）\nfrom connectors.shopify_connector import ProductPayload, VariantPayload, ImagePayload, SEOPayload\nproduct = ProductPayload(\n    title=\"Premium Widget\",\n    body_html=\"<p>High quality widget</p>\",\n    vendor=\"Acme Corp\",\n    product_type=\"Electronics\",\n    status=\"active\",\n    tags=[\"widget\", \"premium\"],\n    variants=[\n        VariantPayload(\n            title=\"Blue / Large\",\n            price=\"29.99\",\n            sku=\"WGT-BL-L\",\n            inventory_quantity=100,\n            option1=\"Blue\", option2=\"Large\",\n        ),\n        VariantPayload(\n            title=\"Red / Large\",\n            price=\"29.99\",\n            sku=\"WGT-RE-L\",\n            inventory_quantity=50,\n            option1=\"Red\", option2=\"Large\",\n        ),\n    ],\n    images=[\n        ImagePayload(src=\"https://cdn.example.com/widget.jpg\", alt_text=\"Widget front view\"),\n    ],\n    seo=SEOPayload(title=\"Premium Widget - Acme Corp\", description=\"Shop premium widgets\"),\n)\n# 通过内部 GraphQL API 直接创建\nresp = shopify._graphql(shopify._build_product_create_mutation(product))\n\n# ── 变体操作 ──────────────────────────────────────────\nvariant = VariantPayload(\n    title=\"Green / Medium\",\n    price=\"24.99\",\n    sku=\"WGT-GN-M\",\n    inventory_quantity=30,\n)\nshopify.create_variant(product_id=123, variant=variant)\nshopify.update_variant(variant_id=555, variant=VariantPayload(price=\"22.99\", inventory_quantity=10))\nshopify.delete_variant(variant_id=555)\n\n# ── 图片操作 ─────────────────────────────────────────\nfrom connectors.shopify_connector import ImagePayload\nshopify.add_product_image(123, ImagePayload(src=\"https://cdn.example.com/extra.jpg\"))\nshopify.delete_product_image(image_id=\"111\")\n\n# ── 商品列表 ─────────────────────────────────────────\nproducts = shopify.list_content({\"first\": 50, \"query\": \"status:active\"})\nfor p in products:\n    print(f\"  {p['id']} | {p['title']} | {p['handle']}\")\n\n# ── 删除商品（归档） ──────────────────────────────────\nshopify.delete_content(product_id=123)       # 归档（soft delete）\nshopify.delete_content(product_id=123, force=True)  # 永久删除\n\n# ── 历史 ─────────────────────────────────────────────\nhistory = shopify.get_history()\nfor h in history:\n    print(f\"  [{h['timestamp']}] {h['operation']} - {h['status']}\")\n\n# ── 回滚 ─────────────────────────────────────────────\n# 回滚最近一次创建操作\nfor rec in reversed(history):\n    if rec[\"operation\"] == \"create\" and rec[\"status\"] == \"executed\":\n        shopify.rollback_operation(rec[\"id\"])\n        print(\"✅ 回滚成功\")\n        break\n\nshopify.close()\n```\n\n---\n\n## 2. Amazon SP-API 连接器\n\n### 认证配置\n\n1. 在 Seller Central 注册 SP-API 应用，获取 Client ID 和 Client Secret\n2. 完成 LWA OAuth 流程，获取 Refresh Token\n3. Refresh Token 可长期有效，用于自动刷新 Access Token\n\n### 基本使用\n\n```python\nfrom connectors import AmazonConnector, CMSCredential, ContentPayload\n\n# 初始化连接器\ncred = CMSCredential(\n    url=\"https://sellercentral.amazon.com\",\n    username=\"amzn1.application-xxx.client_id\",\n    app_password=\"amzn1.application-xxx.client_secret\",\n    api_key=\"Atza|xxx-access-token\",  # 可选，已有 access token\n)\namazon = AmazonConnector(\n    cred,\n    region=\"na\",                    # na | eu | fe\n    marketplace_id=\"A1AM79NJPZON8\",  # 美国市场\n)\n\n# 设置 Refresh Token（用于自动续期 access token）\namazon.set_refresh_token(\"Atzr|...\")\n\n# 验证连接\nif amazon.authenticate():\n    print(\"✅ Amazon SP-API 连接成功\")\n\n# ── 商品列表操作 ─────────────────────────────────────\nfrom connectors.amazon_connector import AmazonListingPayload\n\n# 方式一：使用 ContentPayload\npayload = ContentPayload(title=\"USB-C Cable\", content=\"High-speed USB-C cable 2m\")\nresult = amazon.create_content(payload)\n\n# 方式二：使用 AmazonListingPayload（完整字段）\nlisting = AmazonListingPayload(\n    sku=\"USB-C-2M-BLK\",\n    product_type=\"ELECTRONIC_ACCESSORY\",\n    title=\"USB-C to USB-C Cable 2m Fast Charging\",\n    description=\"Braided nylon, 100W PD fast charge\",\n    brand=\"CableMax\",\n    manufacturer=\"CableMax Inc\",\n    price_amount=12.99,\n    price_currency=\"USD\",\n    quantity=500,\n    condition_type=\"New\",\n    fulfillment_channel=\"MFN\",\n    bullet_points=[\n        \"100W Power Delivery\",\n        \"USB 3.1 Gen 2, 10Gbps\",\n        \"Durable braided nylon\",\n    ],\n)\nresp = amazon._spapi_request(\n    \"PUT\",\n    f\"/listings/2021-08-01/items/{SELLER_ID}/{listing.sku}?marketplaceId={marketplace}\",\n    token=access_token,\n    data={\"attributes\": {...}},\n)\n\n# 更新商品\nresult = amazon.update_content(sku_int, payload)\n\n# 删除商品\nresult = amazon.delete_content(sku_int)\n\n# ── 库存管理 ──────────────────────────────────────────\n# 更新库存\namazon.update_inventory(sku=\"USB-C-2M-BLK\", quantity=300, fulfillment_channel=\"MFN\")\n# 查询库存\ninv = amazon.get_inventory(\"USB-C-2M-BLK\")\n# 批量查询\ninventories = amazon.list_inventory({\"next_token\": \"\"})\n\n# ── 价格管理 ──────────────────────────────────────────\n# 更新价格\namazon.update_pricing(sku=\"USB-C-2M-BLK\", amount=14.99, currency=\"USD\")\n# 查询价格\nprice = amazon.get_pricing(\"USB-C-2M-BLK\")\n\n# ── 报告 ─────────────────────────────────────────────\nfrom connectors.amazon_connector import PricePayload, InventoryPayload\n\n# 请求商品报告\nreport = amazon.request_report(\n    report_type=\"GET_MERCHANT_LISTINGS_DATA\",\n    marketplace_ids=[\"A1AM79NJPZON8\"],\n)\nprint(f\"Report ID: {report['report_id']}, Status: {report['status']}\")\n\n# 查询报告状态\nstatus = amazon.get_report(report[\"report_id\"])\n# 获取报告下载链接\ndocument = amazon.get_report_document(status[\"payload\"][\"reportDocumentId\"])\ndownload_url = document[\"url\"]  # 预签名 S3 URL\n\n# ── Feed 提交（批量 XML） ─────────────────────────────\nxml_content = '<?xml version=\"1.0\"?><AmazonEnvelope><Header>...</Header></AmazonEnvelope>'\nfeed = amazon.submit_feed(\n    feed_type=\"POST_PRODUCT_DATA\",\n    content=xml_content,\n    content_type=\"text/xml\",\n)\nprint(f\"Feed ID: {feed['feed_id']}, Status: {feed['status']}\")\n\n# ── 历史与回滚 ────────────────────────────────────────\nhistory = amazon.get_history()\nfor h in history:\n    print(f\"  [{h['timestamp']}] {h['operation']} {h['entity_type']} - {h['status']}\")\n\namazon.close()\n```\n\n---\n\n## 3. 审批流程（通用）\n\n```python\nfrom engine import ApprovalEngine, ApprovalRule, ApprovalLevel, OperationType\n\nengine = ApprovalEngine()\n\n# 创建审批请求（草稿 → 自动通过）\nreq = engine.create_request(\n    submitter=\"auto-bot\",\n    operation=OperationType.CREATE,\n    payload={\"title\": \"新商品\", \"content\": \"...\", \"status\": \"draft\"},\n)\n\nif engine.can_auto_approve(req):\n    engine.auto_approve(req)\n    print(\"✅ 自动审批通过\")\nelse:\n    print(f\"⏳ 等待审批，级别: {req.level.value}\")\n    engine.vote(req.id, approver=\"admin\", decision=\"approve\")\n```\n\n---\n\n## 4. 回滚操作\n\n```python\nfrom engine import RollbackEngine\n\nrollback = RollbackEngine(shopify)  # 或 AmazonConnector 实例\n\n# 查看最近变更\nchanges = rollback.list_recent_changes(hours=24)\nfor c in changes:\n    print(f\"  {c['id']} | {c['operation']} | {c['status']}\")\n\n# 生成回滚计划（不执行）\nplan = rollback.plan_rollback(record_id=\"abc12345\")\nprint(f\"回滚计划: {plan.strategy.value}, 步骤数: {len(plan.steps)}\")\n\n# 执行回滚\nresult = rollback.execute_rollback(plan.plan_id, force=True)\nprint(f\"回滚结果: {'成功' if result['success'] else '失败'}\")\n```\n\n---\n\n## 风险分级\n\n| 操作 | 触发条件 | 审批级别 |\n|------|---------|---------|\n| 创建商品 | 草稿状态 + 内容安全 | **AUTO_PASS** |\n| 更新商品 | 任意状态 | **SINGLE** |\n| 删除商品 | 任意状态 | **MULTI_SIGN** |\n| 价格变更 | 价格修改 | **MULTI_SIGN** |\n| 包含危险关键词 | `<script>`/`<?php` 等 | **MULTI_SIGN** |\n\n**AUTO_PASS**：无需审批，系统自动通过\n**SINGLE**：单人审批（任意一位审批人同意即可）\n**MULTI_SIGN**：多人会签（所有审批人全部同意）\n**ANY_SIGN**：任意一人同意即可\n\n---\n\n## 测试运行\n\n```bash\ncd agent-cluster/cms-executor\n\n# 运行 Shopify & Amazon 连接器测试（推荐）\npython -m pytest tests/test_connectors.py -v\n\n# 运行 WordPress & 审批引擎测试\npython -m pytest tests/test_cms_executor.py -v\n\n# 运行全部测试\npython -m pytest tests/ -v\n\n# 或直接运行\npython tests/test_connectors.py\npython tests/test_cms_executor.py\n```\n\n---\n\n## 扩展其他 CMS\n\n继承 `BaseCMSConnector` 抽象类，实现以下方法即可：\n\n```python\nfrom connectors.base_connector import BaseCMSConnector, CMSCredential, ContentPayload\n\nclass MyCMSConnector(BaseCMSConnector):\n    def authenticate(self) -> bool: ...\n    def create_content(self, payload: ContentPayload) -> Dict[str, Any]: ...\n    def update_content(self, content_id: int, payload: ContentPayload) -> Dict[str, Any]: ...\n    def delete_content(self, content_id: int, force: bool = False) -> Dict[str, Any]: ...\n    def get_content(self, content_id: int) -> Dict[str, Any]: ...\n    def list_content(self, params: Dict[str, Any] = None) -> List[Dict[str, Any]]: ...\n    def _do_rollback(self, record: OperationRecord) -> bool: ...\n```\n\nFile v3.0.4:README.md\n\n# 企业级智能体集群系统 v3.0\n\n> 基于 **1+N** 架构的智能体协作系统，参考 OpenClaw Main Agent、腾讯ADP Router、智己汽车研发设计集群\n\n## 核心升级（v3.0）\n\n本次更新将执行层解耦，支持**三引擎热切换**：\n\n1. **执行层抽象（ExecutionEngine）**：统一接口抽象，支持 Claude MA / Local / DeepSeek 三种引擎\n2. **EngineRouter 智能路由**：基于意图/场景/关键词的自动引擎选择，支持降级\n3. **向后兼容**：现有 `handle_request` API 完全不受影响，新增 `execute_with_engine` API\n\n## 核心升级（v2.0）\n\n本次更新根据用户评测反馈进行了三大改进：\n\n1. **真实API接入层**：适配器模式支持多ERP系统接入（SAP/用友/金蝶等），保留模拟数据用于开发/演示\n2. **跨Agent协作流程细化**：细粒度任务协议、状态同步、链路追踪、并行/串行混合执行引擎\n3. **错误处理和状态管理**：统一异常中间件、任务状态机（pending→running→success/failed/retry）、重试策略、操作日志\n\n## 系统架构\n\n```\n┌─────────────────────────────────────────────────────────────┐\n│                     用户请求 (自然语言)                       │\n└───────────────────────┬─────────────────────────────────────┘\n                        │\n                        ▼\n┌─────────────────────────────────────────────────────────────┐\n│              Orchestrator (指挥智能体) v3.0                   │\n│  ┌─────────────┐ ┌─────────────┐ ┌──────────────────────┐  │\n│  │ 意图识别    │→│ 任务拆解    │→│ 智能体调度 (串行/并行) │  │\n│  └─────────────┘ └─────────────┘ └──────────────────────┘  │\n│  ┌──────────────────────────────────────────────────────┐  │\n│  │  ExecutionLayer: EngineRouter（三引擎热切换）           │  │\n│  │  ┌────────────┐ ┌─────────────┐ ┌──────────────────┐  │  │\n│  │  │ClaudeMA    │ │LocalEngine  │ │DeepSeekEngine    │  │  │\n│  │  │(通用任务)  │ │(垂直知识)   │ │(合规场景)        │  │  │\n│  │  └────────────┘ └─────────────┘ └──────────────────┘  │  │\n│  └──────────────────────────────────────────────────────┘  │\n└───────────────────────┬─────────────────────────────────────┘\n                        │\n        ┌───────────────┼───────────────┬───────────────┐\n        ▼               ▼               ▼               ▼\n┌───────────────┐ ┌─────────────┐ ┌───────────────┐ ┌───────────────┐\n│Inventory Agent│ │Logistics    │ │Procurement    │ │Finance Agent  │\n│(库存智能体)   │ │Agent        │ │Agent          │ │(财务智能体)   │\n│               │ │(物流智能体) │ │(采购智能体)   │ │               │\n└───────┬───────┘ └──────┬──────┘ └───────┬───────┘ └───────┬───────┘\n        │                 │                 │                 │\n        ▼                 ▼                 ▼                 ▼\n┌─────────────────────────────────────────────────────────────┐\n│          API Integration Layer（v2.0）                        │\n│  ┌──────────────┐  ┌──────────────┐  ┌──────────────────┐   │\n│  │ SAP适配器    │  │ 用友适配器   │  │ 健康监控+断路器  │   │\n│  │ 金蝶适配器   │  │ 通用REST    │  │ 故障自动降级     │   │\n│  └──────────────┘  └──────────────┘  └──────────────────┘   │\n└─────────────────────────────────────────────────────────────┘\n```\n\n## 执行引擎对比\n\n| 维度 | ClaudeMAEngine | LocalEngine | DeepSeekEngine |\n|------|---------------|-------------|----------------|\n| 场景 | 通用开发任务 | 垂直行业任务 | 合规场景 |\n| 凭证管理 | ✅ 官方托管 | ✅ 完全本地 | ✅ 需配置 |\n| 自进化 | ❌ | ✅ M-A3独有 | ❌ |\n| 垂直知识 | ❌ 需自建 | ✅ 塑化行业 | ❌ |\n| 合规认证 | ❌ | ✅ 完全离线 | ✅ 国产合规 |\n| 流式输出 | ✅ | ✅ | ✅ |\n| 多模态 | ✅ | ❌ | ❌ |\n\n## 目录结构\n\n```\nagent-cluster/\n├── __init__.py                   # 包入口（v3.0）\n│\n├── orchestrator.py               # 指挥智能体（核心调度器）v3.0 新增 execute_with_engine\n├── README.md                     # 本文件\n│\n├── execution/                    # 【新增 v3.0】执行引擎抽象层\n│   ├── __init__.py              # 包入口，统一导出\n│   ├── engine_base.py          # ExecutionEngine 抽象基类 + ExecutionResult\n│   ├── engine_router.py         # EngineRouter 路由器 + RoutingRule/RoutingContext\n│   ├── claude_ma_engine.py     # Claude Managed Agents 适配器\n│   ├── local_engine.py         # 本地自建引擎（Orchestrator 原有逻辑迁移）\n│   └── deepseek_engine.py      # 国产模型适配器（DeepSeek/华为等）\n│\n├── api_integration/              # 【v2.0】真实API接入层\n│   ├── __init__.py\n│   ├── api_adapter.py           # 多ERP适配器（SAP/用友/金蝶）\n│   ├── api_config.py            # 配置化管理（环境变量/YAML）\n│   ├── api_health.py            # 健康检查+断路器+告警\n│   └── mock_data.py            # 模拟数据（开发/演示模式）\n│\n├── collaboration/                # 【新增】跨Agent协作流程\n│   ├── __init__.py\n│   ├── task_protocol.py         # 细粒度任务协议（TaskMessage）\n│   ├── state_sync.py            # Agent间状态同步+TTL+订阅通知\n│   ├── trace_tracker.py         # 协作链路追踪+Mermaid时序图\n│   └── workflow_engine.py        # 混合执行引擎（串行/并行/混合）\n│\n├── error_handling/              # 【新增】错误处理与状态管理\n│   ├── __init__.py\n│   ├── task_state_machine.py    # 任务状态机（7种状态+转换规则）\n│   ├── exception_middleware.py  # 统一异常处理+分类+告警\n│   ├── retry_policy.py          # 重试策略（5种）+条件重试\n│   └── operation_log.py        # 操作日志+脱敏+合规报告\n│\n├── specialists/                 # 专业智能体\n│   ├── inventory_agent.py      # 库存智能体\n│   ├── logistics_agent.py       # 物流智能体\n│   ├── procurement_agent.py     # 采购智能体\n│   ├── finance_agent.py         # 财务智能体\n│   └── doc_agent.py            # 工艺文档智能体\n│\n├── mcp_servers/                 # MCP协议封装\n│   ├── erp_server.py           # ERP系统接口\n│   ├── wms_server.py           # WMS仓库管理接口\n│   └── srm_server.py           # SRM供应商管理接口\n│\n├── safety/                      # 安全围栏\n│   ├── permission_manager.py    # RBAC权限管理\n│   ├── audit_logger.py          # 全链路审计日志\n│   └── human_loop.py            # 人机回环审批\n│\n├── config/                     # 配置文件\n│   ├── agents.yaml             # 智能体定义\n│   ├── workflows.yaml          # 工作流配置\n│   ├── permissions.yaml        # 权限矩阵\n│   └── engines.yaml            # 【新增 v3.0】引擎路由规则配置\n│\n└── tests/                     # 【新增 v3.0】单元测试\n    └── test_execution_engines.py  # 执行引擎测试套件\n```\n\n## 快速开始\n\n### 环境要求\n\n- Python 3.10+\n- 依赖包：\n  ```bash\n  pip install pyyaml fastapi uvicorn httpx aiofiles\n  ```\n\n### 运行演示\n\n```bash\n# 完整演示（指挥智能体）\ncd agent-cluster\npython orchestrator.py\n\n# 单独测试各智能体\npython -m specialists.inventory_agent\npython -m specialists.procurement_agent\npython -m specialists.finance_agent\n```\n\n---\n\n## API接入层详解（v2.0新增）\n\n### 支持的ERP类型\n\n| ERP类型 | 适配器 | API协议 | 配置示例 |\n|---------|--------|---------|---------|\n| SAP S/4HANA | `SAPERPAdapter` | OData/REST | `SAP_BASE_URL` |\n| 用友U8/NC/YonBIP | `YonyouERPAdapter` | REST API | `YONYOU_BASE_URL` |\n| 金蝶K3 Cloud/EAS | `KingdeeERPAdapter`（可扩展） | REST API | `KINGDEE_BASE_URL` |\n| 通用REST | `CustomRESTAdapter`（可扩展） | OpenAPI | `CUSTOM_BASE_URL` |\n| 模拟模式 | `MockDataGenerator` | - | `SYSTEM_MODE=demo` |\n\n### 配置方式\n\n**方式1：环境变量**\n```bash\nexport SYSTEM_MODE=demo          # demo/production/development\nexport SAP_BASE_URL=https://sap.example.com\nexport SAP_API_KEY=sk-xxx\nexport YONYOU_BASE_URL=https://yonyou.example.com\nexport YONYOU_APPKEY=your_appkey\n```\n\n**方式2：YAML配置文件**\n```yaml\nsystem:\n  mode: demo\n  log_level: INFO\n  enable_trace: true\n  demo_variance: 0.1\n\nerp_systems:\n  - name: sap_primary\n    erp_type: sap\n    is_primary: true\n    base_url: https://sap.example.com\n    auth_type: bearer\n    api_key: ${SAP_API_KEY}\n    timeout: 30.0\n    circuit_breaker_threshold: 5\n```\n\n### 代码示例\n\n```python\nfrom api_integration import APIConfigManager, MockDataGenerator\n\n# 自动从环境变量加载配置\nconfig = APIConfigManager.from_env()\n\n# 演示模式使用模拟数据\nif config.system.mode == \"demo\":\n    mock = MockDataGenerator(variance=0.1)\n    response = mock.query_inventory(sku=\"SKU001\")\n    print(response.data)\n```\n\n---\n\n## 协作流程详解（v2.0新增）\n\n### 细粒度任务协议\n\n```python\nfrom collaboration import TaskMessage, TaskContext, TaskPriority, TaskMode\n\n# 创建任务消息\ntask = TaskMessage(\n    agent_name=\"inventory_agent\",\n    action=\"query_stock\",\n    parameters={\"sku\": \"SKU001\", \"warehouse\": \"华东仓\"},\n    priority=TaskPriority.HIGH,\n    mode=TaskMode.SERIAL,\n    timeout_seconds=30.0,\n    max_retries=3,\n    dependency=TaskDependency(\n        depends_on=[\"task_001\", \"task_002\"],  # 依赖的任务ID\n        blocking=True,\n        shared_context=[\"budget_summary\"],     # 需要共享的上下文\n    ),\n    context=TaskContext(request_id=\"REQ001\", trace_id=\"trace_xxx\"),\n)\n```\n\n### 状态同步\n\n```python\nfrom collaboration import SharedStateManager\n\nstate = SharedStateManager(agent_id=\"inventory_agent\")\n\n# 设置状态（TTL=300秒）\nawait state.set(\"inventory:SKU001\", {\"qty\": 450}, ttl_seconds=300)\n\n# 订阅变更\nstate.subscribe(\n    \"inventory:*\",\n    lambda key, value, entry: print(f\"库存更新: {key}={value}\"),\n    subscriber=\"logistics_agent\",\n)\n```\n\n### 链路追踪（Mermaid时序图）\n\n```python\nfrom collaboration import CollaborationTracker\n\ntracker = CollaborationTracker()\ntrace_id = tracker.start_trace(\"REQ001\", user_input=\"查询库存\")\n\n# 执行任务并记录\nspan = tracker.start_span(trace_id, \"inventory_agent:query_stock\", SpanType.AGENT)\n# ... 执行逻辑 ...\ntracker.end_span(span, status=\"ok\")\n\n# 导出Mermaid时序图\nprint(tracker.to_mermaid_sequence(trace_id))\n```\n\n输出示例：\n```mermaid\nsequenceDiagram\n    participant 用户\n    participant inventory_agent\n    participant erp_system\n    用户->>+inventory_agent: query_stock(SKU001)\n    inventory_agent->>+erp_system: GET /api/stock\n    erp_system-->>-inventory_agent: {qty: 450}\n    inventory_agent-->>-用户: {status: ok}\n```\n\n---\n\n## 错误处理详解（v2.0新增）\n\n### 状态机\n\n```python\nfrom error_handling import TaskStateMachine, State\n\nsm = TaskStateMachine(auto_retry=True, default_max_retries=3)\n\n# 创建任务\ntask = sm.create_task(\"task_001\", \"查询库存\", \"inventory_agent\")\n\n# 状态转换（自动校验合法性）\nsm.start(\"task_001\")         # pending → running\nsm.succeed(\"task_001\", result={\"qty\": 450})  # → success\n\n# 失败时自动重试（指数退避）\nsm.fail(\"task_001\", \"网络超时\")\n# → running → retry（第1次，1s后）→ running → ...\n# → running → retry（第3次，8s后）→ running → ...\n# → failed（超过最大重试次数）\n```\n\n### 统一异常处理\n\n```python\nfrom error_handling import ExceptionMiddleware, handle_exceptions\n\nmiddleware = ExceptionMiddleware()\n\ntry:\n    # ERP API调用\n    response = await adapter.query_inventory(sku=\"SKU001\")\nexcept Exception as e:\n    error = middleware.handle(e, source=\"inventory_agent\", request_id=\"REQ001\")\n    print(error.to_dict())\n    # {\n    #   \"error_id\": \"a1b2c3d4\",\n    #   \"category\": \"network\",\n    #   \"severity\": \"high\",\n    #   \"message\": \"无法连接到ERP系统\",\n    #   \"suggestion\": \"请检查网络连接，ERP系统是否可达，可尝试重试\",\n    #   \"retry_recommended\": true\n    # }\n```\n\n### 重试策略\n\n```python\nfrom error_handling import RetryExecutor, RetryConfig, RetryStrategy\n\nconfig = RetryConfig(\n    max_attempts=5,\n    strategy=RetryStrategy.EXP_JITTER,  # AWS推荐：指数+抖动\n    base_delay=1.0,\n    max_delay=60.0,\n)\n\nexecutor = RetryExecutor(config)\nresult = await executor.execute(\n    lambda: call_erp_api(),\n    on_retry=lambda attempt, exc: alert(f\"重试第{attempt}次: {exc}\"),\n)\n```\n\n### 操作日志\n\n```python\nfrom error_handling import OperationLogger\n\nlogger = OperationLogger(log_dir=\"logs\", compress=True)\n\nlogger.agent_start(\"inventory_agent\", \"REQ001\", trace_id=\"trace_xxx\")\nlogger.agent_end(\"inventory_agent\", \"REQ001\", duration_ms=230.5, success=True)\n\n# 生成合规报告\nreport = logger.generate_report(start_date=datetime(2026, 4, 1))\nprint(report)\n```\n\n---\n\n## 错误处理和状态管理详解\n\n### 状态转换规则\n\n```\nCREATED → PENDING → RUNNING → SUCCESS\n                      ├→ FAILED → RETRY → RUNNING（最多N次）\n                      ├→ TIMEOUT → RETRY → RUNNING（最多N次）\n                      └→ CANCELLED（终态）\n```\n\n### 错误分类\n\n| 分类 | 说明 | 是否可重试 |\n|------|------|-----------|\n| `validation` | 参数校验错误 | ❌ |\n| `network` | 网络/连接错误 | ✅ |\n| `timeout` | 超时错误 | ✅ |\n| `auth` | 认证/权限错误 | ❌ |\n| `not_found` | 资源不存在 | ❌ |\n| `external` | 外部依赖错误 | ✅ |\n| `internal` | 内部错误 | ❌ |\n\n---\n\n## 核心模块详解\n\n### 指挥智能体 (orchestrator.py)\n\n**职责**：不直接干活，只做调度\n\n```\n用户输入 → 意图识别 → 任务拆解 → 智能体分发 → 结果汇总 → 返回\n```\n\n**意图识别示例**：\n\n| 用户输入 | 识别意图 | 调度智能体 |\n|----------|----------|-----------|\n| \"SKU001还有多少库存\" | stock_query | inventory_agent |\n| \"帮我采购一批传感器\" | purchase | procurement_agent + finance_agent |\n| \"货物到哪了\" | logistics | logistics_agent |\n| \"华东仓库存低于安全水位\" | replenishment | inventory + procurement + finance |\n| \"采购+物流全程跟踪\" | procurement_with_logistics | procurement + logistics + finance |\n\n### 专业智能体职责\n\n| 智能体 | 职责 | 关键词 |\n|--------|------|--------|\n| `inventory_agent` | 库存查询、安全水位计算、补货建议 | 库存/安全水位/补货 |\n| `procurement_agent` | 供应商查询、询价、下单跟踪 | 采购/供应商/下单 |\n| `logistics_agent` | 物流计划、运费计算、轨迹追踪 | 物流/运费/发货 |\n| `finance_agent` | 预算查询、付款审核、成本分析 | 财务/预算/付款 |\n| `doc_agent` | 工艺文档、BOM表生成 | 文档/工艺/BOM |\n\n---\n\n## ⭐ 评测与反馈\n\n如果你体验过本系统，欢迎在 **虾评（XiPing）** 平台留下你的评测。\n\n> 本技能目前处于众测期（⭐ 4.3/5.0，已有 3 条评测），需要 **5 条有效评测（≥4分）** 才能转正为正式版。你的每一份反馈都在帮助这个项目变得更好！\n\n👉 **评测链接**：[产业互联网硅基军团 - 虾评](https://xiaping.coze.site/skill/e58e62c8-789d-451c-a009-0cfa89253149)\n\n### 评测维度参考\n\n| 维度 | 说明 |\n|------|------|\n| 功能完整性 | ERP适配器、Multi-Agent协作、错误处理是否如描述工作？ |\n| 代码质量 | 架构设计、模块划分、注释清晰度如何？ |\n| 文档质量 | README、快速开始、API文档是否足够？ |\n| 实用性 | 你的业务场景能否从中受益？ |\n| 改进建议 | 哪些功能你最希望看到加强？ |\n\n### 快速体验建议\n\n```bash\n# 1. 克隆项目\ngit clone https://github.com/your-org/agent-cluster.git\ncd agent-cluster\n\n# 2. 安装依赖\npip install pyyaml fastapi uvicorn httpx aiofiles\n\n# 3. 启动演示（默认 Demo 模式，无需真实 ERP）\npython orchestrator.py\n```\n\n体验后扫码评测，感谢你的支持！ 🙏\n\nFile v3.0.4:_meta.json\n\n{\n  \"ownerId\": \"kn7ch74w4kf43pffbq4pxda0w584hy2v\",\n  \"slug\": \"agent-cluster\",\n  \"version\": \"3.0.4\",\n  \"publishedAt\": 1777647819752\n}\n\nFile v3.0.4:30-expansion/30-agents-design.md\n\n# M-A3 Agent集群扩展方案：从6个到30个\n> 版本：v1.0 | 日期：2026-04-14 | 作者：M-A3 幕僚长\n\n---\n\n## 一、项目概述\n\n### 1.1 背景与目标\n\n当前 M-A3 集群已实现：\n- **geo-ops-agents**：6个Agent三层架构（市场研究→内容策略→效果监测）\n- **amazon-ops-agents**：多个运营Agent并行协作\n- **硅基军团**：20个产业Agent（库存/物流/采购/财务/生产/销售）\n\n**扩展目标**：\n> 构建全球最大的垂直领域Agent集群（30个专业Agent），覆盖**出海GEO营销**、**亚马逊全链路运营**、**平台基础设施支撑**三大功能域，形成代差级竞争优势。\n\n### 1.2 规模对比\n\n| 维度 | 现有水平 | 扩展后 | 提升倍数 |\n|------|---------|--------|---------|\n| Agent数量 | 6 | 30 | 5× |\n| 功能域 | 1 | 3 | 3域 |\n| 协作模式 | 串行为主 | 并行+串行混合 | 质变 |\n| 覆盖阶段 | 单点 | 全链路 | 全链路 |\n\n### 1.3 三层架构（扩展版）\n\n```\n┌─────────────────────────────────────────────────────────────────┐\n│                    幕僚长 M-A3（Chief of Staff）                    │\n│  意图识别 → 任务拆解 → 智能调度（并行/串行）→ 结果聚合 → 报告生成      │\n└───────────────────────────────┬─────────────────────────────────┘\n                                │ 并行/串行混合调度\n        ┌───────────────────────┼───────────────────────┐\n        ▼                       ▼                       ▼\n┌───────────────┐      ┌───────────────┐      ┌───────────────┐\n│   GEO域       │      │  亚马逊域     │      │  支撑域        │\n│  (10 Agents)  │      │  (10 Agents)  │      │  (10 Agents)   │\n│   第2层       │      │   第2层        │      │   第2层        │\n└───────────────┘      └───────────────┘      └───────────────┘\n        │                       │                       │\n        ▼                       ▼                       ▼\n┌─────────────────────────────────────────────────────────────────┐\n│               外部工具层（第1层：Tool/Plugin）                      │\n│  搜索引擎 / 平台API / 浏览器 / 数据库 / 文件系统 / 邮件系统           │\n└─────────────────────────────────────────────────────────────────┘\n```\n\n---\n\n## 二、Agent详细设计\n\n### 2.1 GEO域（10个）— 出海营销与AI搜索优化\n\n#### 🔍 GEO-01 市场研究Agent\n```\n职责：输入产品/品类 → 输出目标市场机会分析报告\n核心能力：\n  - 全球主要市场（北美/欧盟/东南亚/拉美/中东）容量估算\n  - 目标客群画像生成（年龄/收入/购买习惯/文化偏好）\n  - 市场规模计算（TAM/SAM/SOM三层模型）\n  - 季节性需求分析\n  - 监管政策扫描（GDPR/REACH/FCC等合规要求）\n技能包：geo-operations-assistant, search_web, 数据分析\n输入：产品描述、目标市场列表\n输出：市场机会报告（JSON + 可视化建议）\n关键词触发：[\"市场研究\", \"市场规模\", \"目标市场\", \"TAM\", \"市场机会\"]\n优先级：P0（入口节点）\n```\n\n#### 📊 GEO-02 竞品分析Agent\n```\n职责：深度分析竞争对手的GEO策略，找出攻防机会\n核心能力：\n  - 竞品识别（LLM Brand Detection + 搜索结果验证）\n  - GEO三维度评分：内容权威性×技术SEO×品牌信号\n  - 竞品内容策略拆解（话题/格式/渠道/频率）\n  - 竞品反向链接分析\n  - AI搜索引用率对比（Perplexity/SearchGPT/DeepSearch）\n  - Gap分析：竞品覆盖但我方空白的GEO机会点\n技能包：search_web, 数据分析, 知识图谱\n输入：竞争对手列表/产品关键词\n输出：竞品GEO画像 + 攻防策略建议\n关键词触发：[\"竞品分析\", \"竞争对手\", \"竞品策略\", \"对标分析\"]\n优先级：P1\n```\n\n#### ✍️ GEO-03 内容策略Agent\n```\n职责：制定全渠道内容日历和内容策略\n核心能力：\n  - 话题发现（高频问题/长尾疑问/Trending话题）\n  - 内容形式规划（博客/白皮书/视频脚本/社交帖）\n  - 渠道分布策略（官网/知乎/CSDN/Medium/LinkedIn）\n  - 发布频率优化（基于竞品数据+搜索信号）\n  - 内容复用矩阵（一篇长文→多条短帖→多语言版本）\n  - E-E-A-T 信号植入策略\n技能包：内容生成, 翻译, search_web\n输入：产品信息、目标关键词、渠道偏好\n输出：季度内容日历（CSV）+ 话题优先级排序\n关键词触发：[\"内容策略\", \"内容规划\", \"选题\", \"内容日历\"]\n优先级：P0（核心输出节点）\n```\n\n#### 🌐 GEO-04 多语言优化Agent\n```\n职责：实现GEO内容的全球化与本地化\n核心能力：\n  - 语言市场优先级排序（基于搜索量+购买力）\n  - 本地化关键词研究（文化差异+本地搜索引擎差异）\n  - 地道表达生成（避免机翻感）\n  - 文化适配（节日/习俗/禁忌词）\n  - hreflang标签策略\n  - 本地化内容质量评估（Native Speaker风格评分）\n技能包：翻译, search_web, 内容生成\n输入：原语言内容、目标语言列表\n输出：本地化内容 + hreflang配置建议\n关键词触发：[\"多语言\", \"本地化\", \"翻译\", \"国际化\", \"中译英\"]\n优先级：P1\n```\n\n#### 📱 GEO-05 平台适配Agent\n```\n职责：将内容策略适配到各平台的具体要求\n核心能力：\n  - 平台特性分析（Google/Bing/Perplexity/知乎/百度）\n  - 平台内容规范（字数/格式/标签/分类）\n  - AI搜索友好内容格式（结构化数据/FAQ/摘要前置）\n  - 各平台Schema适配\n  - 平台算法偏好分析\n  - 多平台同步发布配置\n技能包：Schema优化, search_web\n输入：原始内容、目标平台列表\n输出：平台适配后的内容 + 发布配置\n关键词触发：[\"平台适配\", \"多平台\", \"SEO适配\", \"平台规则\"]\n优先级：P1\n```\n\n#### 📈 GEO-06 效果监测Agent\n```\n职责：实时追踪GEO策略的执行效果\n核心能力：\n  - AI搜索引用率监控（Perplexity/SearchGPT）\n  - 关键词排名追踪（Google/Bing/百度）\n  - 流量来源分析（GA4适配/自定义事件）\n  - AI搜索转化漏斗（Awares→Consider→Convert）\n  - 异常波动告警（下降/上升）\n  - 周报/月报自动生成\n技能包：数据分析, search_web, 报告生成\n输入：监测关键词列表、时间范围\n输出：效果监测报告 + 优化建议\n关键词触发：[\"效果监测\", \"排名追踪\", \"AI搜索\", \"流量分析\", \"GEO效果\"]\n优先级：P1\n```\n\n#### 🧠 GEO-07 知识图谱Agent\n```\n职责：构建和维护品牌的知识图谱（GEO基础设施）\n核心能力：\n  - Schema.org标准实体识别与抽取\n  - 知识图谱构建（三元组：实体-关系-属性）\n  - JSON-LD结构化数据生成\n  - 知识图谱补全（缺失实体预测）\n  - 多源知识融合（官网/维基/社交媒体）\n  - 知识新鲜度管理\n技能包：知识图谱, JSON处理, 数据分析\n输入：品牌信息、产品数据、企业知识文档\n输出：知识图谱文件（JSON-LD）+ Schema配置\n关键词触发：[\"知识图谱\", \"Schema\", \"实体识别\", \"结构化数据\"]\n优先级：P0（GEO基础设施）\n```\n\n#### 🎯 GEO-08 意图预测Agent\n```\n职责：预测用户搜索意图，指导内容生成方向\n核心能力：\n  - 搜索意图分类（Informational/Navigational/Transactional）\n  - 用户旅程映射（AIDA模型）\n  - 高价值意图词挖掘\n  - 意图随时间/事件的演变分析\n  - 竞品意图覆盖分析\n  - 意图-内容匹配度评分\n技能包：意图识别, 数据分析, search_web\n输入：产品类别、核心关键词\n输出：意图分析矩阵 + 内容匹配建议\n关键词触发：[\"意图预测\", \"搜索意图\", \"用户意图\", \"意图分析\"]\n优先级：P1（参考PureblueAI的94.3%意图预测水平）\n```\n\n#### 🏷️ GEO-09 Schema优化Agent\n```\n职责：持续优化网站的Schema标记，提升AI搜索可见性\n核心能力：\n  - 全站Schema审计（覆盖率/错误率/完整性）\n  - 页面级别Schema推荐（Product/FAQ/Article/Organization）\n  - Rich Results测试与验证\n  - AI搜索信号增强（HowTo/StepByStep结构）\n  - FAQ架构优化（People Also Ask）\n  - 竞品Schema对比分析\n技能包：Schema优化, search_web, 数据分析\n输入：网站URL/内容页面列表\n输出：Schema优化报告 + JSON-LD代码建议\n关键词触发：[\"Schema优化\", \"结构化数据\", \"Rich Results\", \"SEO技术\"]\n优先级：P2（技术增强）\n```\n\n#### 🗺️ GEO-10 地域策略Agent\n```\n职责：制定基于地理位置的差异化GEO策略\n核心能力：\n  - 区域市场特征分析（文化/经济/监管差异）\n  - 本地SEO策略（Google My Business/区域目录）\n  - 区域定价信号优化\n  - 本地化反向链接策略\n  - 区域内容偏好分析\n  - 多区域站点架构建议（ccTLD/subdirectory/subdomain）\n技能包：geo-operations-assistant, 数据分析, search_web\n输入：产品/品牌、目标区域列表\n输出：地域差异化策略报告\n关键词触发：[\"地域策略\", \"本地SEO\", \"区域市场\", \"ccTLD\"]\n优先级：P2（拓展阶段）\n```\n\n---\n\n### 2.2 亚马逊域（10个）— 全链路运营自动化\n\n#### 🛒 AMZ-01 选品分析Agent\n```\n职责：输入市场数据 → 输出选品建议与风险评估\n核心能力：\n  - 市场需求挖掘（关键词搜索量/增长率/季节性）\n  - 竞争度分析（BSR分布/评论数量/评分分布）\n  - 利润空间测算（成本/物流/FBA/平台费/广告）\n  - 差异化机会识别（功能/设计/包装创新）\n  - 合规风险筛查（商标/专利/类目审核）\n  - 生命周期预测（导入期/成长期/成熟期/衰退期）\n技能包：数据分析, search_web, 合规检查\n输入：品类/关键词/预算限制\n输出：选品分析报告（Top10推荐 + 风险评级）\n关键词触发：[\"选品\", \"新品开发\", \"市场调研\", \"产品机会\"]\n优先级：P0（业务入口）\n```\n\n#### 📝 AMZ-02 Listing优化Agent\n```\n职责：打造高转化的亚马逊商品页面\n核心能力：\n  - 标题优化（字符限制/核心词前置/品牌词策略）\n  - 五点描述撰写（痛点-解决方案-证明材料）\n  - 产品描述优化（A+内容结构）\n  - 关键词植入（自然嵌入，避免关键词堆砌）\n  - 图片ALT标签优化\n  - 竞品Listing对比评分\n  - 转化率预测\n技能包：内容生成, 关键词研究, 数据分析\n输入：产品信息、竞品ASIN列表\n输出：完整Listing文档 + 优化建议报告\n关键词触发：[\"Listing优化\", \"标题优化\", \"五点描述\", \"产品页面\"]\n优先级：P0（转化核心）\n```\n\n#### 💰 AMZ-03 利润优化Agent\n```\n职责：最大化每个SKU的净利润\n核心能力：\n  - 全成本建模（FBA费用/仓储费/退款率/广告ACOS）\n  - TACOS（全链路广告成本）优化\n  - 动态定价策略（日内调价/竞品调价响应）\n  - 利润-排名平衡曲线分析\n  - 促销策略优化（Coupon/Deal/LD/7DD）\n  - 冗余库存成本计算与清仓建议\n  - 利润多市场横向对比\n技能包：ProfitOptimizer, 数据分析\n输入：ASIN列表、财务目标（目标ACOS/目标利润率）\n输出：利润优化报告 + 定价建议 + 促销日历\n关键词触发：[\"利润优化\", \"ACOS\", \"TACOS\", \"定价策略\", \"成本分析\"]\n优先级：P0（核心商业指标）\n```\n\n#### 📢 AMZ-04 广告投放Agent\n```\n职责：全托管式广告运营（SP/SB/SD/OTT）\n核心能力：\n  - 关键词挖掘（海量词库+语义扩展）\n  - 广告结构设计（自动广告→手动广告漏斗）\n  - Bid智能调节（基于转化/基于ROAS/日内波动）\n  - 预算分配优化（ Campaigns × Portfolios）\n  - 否定关键词管理\n  - 竞品ASIN定向策略\n  - 广告报告解读与异常诊断\n技能包：广告管理, 数据分析, 关键词研究\n输入：ASIN、预算、广告目标\n输出：广告运营报告 + 下周期调整建议\n关键词触发：[\"广告投放\", \"SP广告\", \"SB广告\", \"ACOS\", \"广告优化\"]\n优先级：P0（流量引擎）\n```\n\n#### 📦 AMZ-05 库存管理Agent\n```\n职责：确保库存充足且无积压\n核心能力：\n  - 补货时间计算（Lead Time × 销售速度 × 安全库存）\n  - 断货风险预警（前置期波动/促销放量）\n  - 冗余库存识别与清仓建议\n  - FBA容量规划（季度容量预测）\n  - 多ASIN库存分配优化\n  - 物流模式选择（海运/空运/快递经济性对比）\n  - 季度性备货规划\n技能包：库存管理, 数据分析\n输入：ASIN列表、在途库存、日销售数据\n输出：补货计划表 + 库存预警报告\n关键词触发：[\"库存管理\", \"补货\", \"FBA库存\", \"库存预警\"]\n优先级：P1（运营保障）\n```\n\n#### ⭐ AMZ-06 评价分析Agent\n```\n职责：监控分析竞品和自己产品的评价\n核心能力：\n  - 全网评价数据采集（亚马逊/独立站/社交媒体）\n  - 评分趋势监控（周环比/异常波动告警）\n  - 评价情感分析（Positive/Negative/Neutral + 细粒度主题）\n  - 竞品弱点挖掘（差评高频词 → 产品改进机会）\n  - 好评模式识别（用于激励计划设计）\n  - 催评策略优化（时间/文案/A+B测试）\n  - QA问题分析（常见问题 → 运营改进）\n技能包：数据分析, 自然语言处理, search_web\n输入：ASIN列表、监控频率设置\n输出：评价分析周报 + 运营行动建议\n关键词触发：[\"评价分析\", \"差评监控\", \"好评优化\", \"review\", \"QA分析\"]\n优先级：P1\n```\n\n#### 👁️ AMZ-07 竞品监控Agent\n```\n职责：实时监控竞品动态，快速响应市场变化\n核心能力：\n  - 竞品价格实时追踪（价格战告警）\n  - 竞品Listing变更检测（标题/图片/五点/价格）\n  - 竞品库存状态监控\n  - 竞品广告策略分析（关键词/出价/广告格式）\n  - 竞品促销日历追踪\n  - 新竞品发现（市场新入局者预警）\n  - 竞品市场份额变化\n技能包：竞品监控, search_web, 数据分析\n输入：竞品ASIN列表、监控维度设置\n输出：竞品动态日报 + 响应策略建议\n关键词触发：[\"竞品监控\", \"价格监控\", \"市场情报\", \"竞争对手动态\"]\n优先级：P1\n```\n\n#### 💵 AMZ-08 定价策略Agent\n```\n职责：智能定价，维持竞争力与利润的动态平衡\n核心能力：\n  - 竞品价格带分析（价格锚定策略）\n  - 动态定价规则引擎（基于库存/排名/竞品/利润）\n  - Buy Box监控与获取策略\n  - 促销活动价格测算（折扣力度 vs 转化率）\n  - 跟卖监控与应对\n  - 新品期/成长期/成熟期差异化定价\n  - 心理定价策略（9.99/19.99等）\n技能包：定价策略, 数据分析, 利润优化\n输入：ASIN、成本结构、竞争环境\n输出：定价策略报告 + 自动调价规则配置\n关键词触发：[\"定价策略\", \"动态定价\", \"价格战\", \"Buy Box\"]\n优先级：P1\n```\n\n#### 🔑 AMZ-09 关键词Agent\n```\n职责：发现、管理和优化所有广告与Listing关键词\n核心能力：\n  - 关键词词库构建（百万级）\n  - 搜索量/竞争度/转化率三维评估\n  - 否定关键词智能推荐\n  - 长尾关键词挖掘\n  - 关键词趋势追踪（新兴词/衰退词）\n  - ASIN关键词反查（竞品流量词）\n  - 关键词分类体系维护（品牌词/类目词/竞品词/长尾词）\n技能包：关键词研究, 数据分析\n输入：ASIN/产品类别/种子关键词\n输出：关键词矩阵 + 优先级排序 + 投放建议\n关键词触发：[\"关键词\", \"keyword\", \"长尾词\", \"搜索词\", \"否定词\"]\n优先级：P1（基础设施）\n```\n\n#### 📊 AMZ-10 报表分析Agent\n```\n职责：自动生成各类业务报表，支持决策\n核心能力：\n  - 日/周/月/季报表自动生成\n  - 核心KPI追踪（GMV/ACOS/FBA库存/BSR/评分）\n  - 同比/环比趋势分析\n  - 异常点标注与归因分析\n  - 多维度下钻（类目/品牌/店铺/区域/时间段）\n  - 自定义报表配置\n  - 可视化图表生成（折线/柱状/热力/漏斗）\n  - 报表推送（邮件/飞书/Slack）\n技能包：报表生成, 数据分析\n输入：报表类型、时间范围、筛选条件\n输出：结构化报表文件（Excel/PDF/HTML）\n关键词触发：[\"报表\", \"数据分析\", \"KPI\", \"周报\", \"月报\", \"业绩分析\"]\n优先级：P2（管理支撑）\n```\n\n---\n\n### 2.3 支撑域（10个）— 平台基础设施与协作底座\n\n#### 🔌 SUP-01 数据采集Agent\n```\n职责：统一数据采集入口，支持所有Agent的数据需求\n核心能力：\n  - 多源数据采集（平台API/网页爬虫/文件导入/数据库）\n  - 数据清洗与标准化\n  - 数据质量校验（完整性/一致性/时效性）\n  - 增量/全量采集配置\n  - 数据管道监控与告警\n  - 数据源健康检查\n技术栈：httpx, BeautifulSoup, pandas, schedule\n输入：数据源配置、数据需求描述\n输出：结构化数据集（JSON/CSV/Pandas DataFrame）\n关键词触发：[\"数据采集\", \"爬虫\", \"数据清洗\", \"数据导入\"]\n优先级：P0（基础设施）\n```\n\n#### ✒️ SUP-02 内容生成Agent\n```\n职责：批量生成高质量营销和运营内容\n核心能力：\n  - 多格式内容生成（产品描述/博客文章/社媒帖子/邮件/视频脚本）\n  - 品牌语调统一（Voice & Tone配置）\n  - 内容模板管理（可复用结构）\n  - 批量内容生成（一次性处理100+条）\n  - A/B内容变体生成\n  - 内容质量评分（可读性/关键词密度/E-E-A-T）\n  - 多语言内容生成（支持20+语言）\n技能包：内容生成, 翻译\n输入：内容需求、产品信息、品牌指南\n输出：批量内容文件 + 质量评分\n关键词触发：[\"内容生成\", \"批量文案\", \"产品描述\", \"脚本生成\"]\n优先级：P0（内容工厂）\n```\n\n#### 🌐 SUP-03 翻译Agent\n```\n职责：提供专业级多语言翻译（不只是翻译，是本地化）\n核心能力：\n  - 20+语言专业翻译（中英日韩德法西意葡俄阿等）\n  - 行业术语库（电商/科技/医疗/金融）\n  - 文化适配（节日/俗语/禁忌）\n  - 翻译质量自评（BLEU辅助，人工审核标记）\n  - 翻译记忆库（TM）复用\n  - 批量翻译任务队列\n  - 紧急翻译通道（24小时加急）\n技能包：翻译, 本地化\n输入：待翻译内容、目标语言、行业领域\n输出：翻译后内容 + 质量评估\n关键词触发：[\"翻译\", \"本地化\", \"中英翻译\", \"多语言\"]\n优先级：P1\n```\n\n#### ✅ SUP-04 合规检查Agent\n```\n职责：确保所有运营行为符合平台政策和法规\n核心能力：\n  - 亚马逊政策检查（禁售商品/受限商品/知识产权）\n  - 广告合规审核（FDA/FTD/极端词汇/竞品提及）\n  - GDPR合规检查（数据收集/用户同意/删除权）\n  - 内容合规扫描（版权/商标/虚假宣传）\n  - 欧盟/美国/中国法规适配\n  - 合规风险评级（Green/Yellow/Red）\n  - 合规报告生成（审计追踪）\n技能包：合规检查, 数据分析\n输入：待检查内容/行为、目标市场/平台\n输出：合规检查报告 + 风险评级\n关键词触发：[\"合规检查\", \"政策合规\", \"风险审核\", \"GDPR\"]\n优先级：P0（风控保障）\n```\n\n#### 📑 SUP-05 报告生成Agent\n```\n职责：自动生成专业级业务报告\n核心能力：\n  - 多格式报告（Markdown/PDF/HTML/PPTX）\n  - 多种报告模板（日报/周报/月报/季报/年报/专项报告）\n  - 数据自动填充（对接所有数据源）\n  - 图表可视化集成（ECharts）\n  - 报告分发（邮件/飞书/Slack/微信）\n  - 报告版本管理（历史版本对比）\n  - 自定义报告配置（拖拽式）\n技能包：报告生成, 数据分析, docx\n输入：报告模板/类型、时间范围、受众\n输出：完整报告文件\n关键词触发：[\"报告生成\", \"报表制作\", \"数据分析报告\", \"PPT\"]\n优先级：P1\n```\n\n#### 💬 SUP-06 客户服务Agent\n```\n职责：处理客户问询，提供智能客服支持\n核心能力：\n  - 意图识别与分类（退款/物流/产品/投诉/咨询）\n  - FAQ自动回复（基于知识库）\n  - 情感分析（识别紧急投诉）\n  - 多轮对话管理\n  - 好评邀请触发（时机判断）\n  - 差评预警与升级机制\n  - 客服工单生成与跟踪\n  - 客服数据统计与分析\n技能包：对话系统, 知识库\n输入：客户问询文本、历史记录\n输出：回复建议/工单/升级建议\n关键词触发：[\"客服\", \"客户问询\", \"自动回复\", \"FAQ\"]\n优先级：P2\n```\n\n#### 🎯 SUP-07 质量评分Agent\n```\n职责：为所有Agent产出提供统一的质量评估\n核心能力：\n  - 多维度质量评分（准确性/完整性/可操作性/时效性）\n  - 质量基线管理（各Agent类型基准分）\n  - 质量趋势追踪\n  - 质量异常告警\n  - 人工抽检机制（随机抽样+人工评估）\n  - 质量改进建议\n  - Agent能力画像（各Agent的强项/弱项）\n技能包：质量评分, 数据分析\n输入：Agent输出内容、任务类型\n输出：质量评分报告 + 改进建议\n关键词触发：[\"质量评分\", \"效果评估\", \"质量监控\"]\n优先级：P1（质量保障层）\n```\n\n#### 🧠 SUP-08 记忆管理Agent\n```\n职责：统一管理集群的长期记忆和知识\n核心能力：\n  - 跨Agent记忆共享（Shared Memory Space）\n  - 记忆层级管理（L1工作记忆/L2会话记忆/L3长期记忆）\n  - 记忆检索（RAG语义搜索）\n  - 记忆去重与合并\n  - 过期记忆处理（自动归档/删除）\n  - 上下文窗口优化（关键信息摘要）\n  - Agent学习经验积累\n技能包：RAG, 记忆管理\n输入：当前上下文、记忆查询\n输出：相关记忆片段 + 更新建议\n关键词触发：[\"记忆管理\", \"知识库\", \"上下文\", \"长期记忆\"]\n优先级：P1（认知基础设施）\n```\n\n#### ⚙️ SUP-09 调度协调Agent\n```\n职责：智能调度所有Agent任务，优化执行效率\n核心能力：\n  - 任务优先级队列管理\n  - 任务依赖图解析（DAG）\n  - 并行任务优化（最大化并发度）\n  - 负载均衡（Agent能力匹配）\n  - 资源配额管理（Token/调用次数/时间）\n  - 任务超时与降级策略\n  - 任务重试与回退机制\n  - 调度策略可视化\n技能包：调度协调, 工作流引擎\n输入：任务列表、约束条件（截止时间/优先级）\n输出：执行计划 + 调度日志\n关键词触发：[\"任务调度\", \"并行执行\", \"工作流\", \"协调\"]\n优先级：P0（执行引擎核心）\n```\n\n#### 🔒 SUP-10 安全审计Agent\n```\n职责：保障整个集群的安全运行\n核心能力：\n  - API密钥安全管理（轮换/吊销/告警）\n  - 权限矩阵维护（RBAC × Agent）\n  - 操作日志审计（可溯源/不可篡改）\n  - 敏感信息检测（PII/信用卡/API Key）\n  - 异常行为检测（频率/范围/时间异常）\n  - SOC 2合规报告生成\n  - MCP协议安全扫描（43%的MCP服务器存在漏洞）\n  - 安全事件响应（自动隔离+告警）\n技能包：安全审计, PII检测\n输入：审计范围（全部/指定Agent/指定时间）\n输出：安全审计报告 + 风险建议\n关键词触发：[\"安全审计\", \"权限管理\", \"日志审计\", \"合规报告\"]\n优先级：P0（安全底线）\n```\n\n---\n\n## 三、协作机制设计\n\n### 3.1 幕僚长调度流程（扩展版）\n\n```\n用户请求\n    ↓\n意图识别（SUP-09 调度协调 + SUP-08 记忆管理）\n    ↓\n任务拆解（幕僚长LLM）\n    ↓\n┌─────────────────────────────────────────────────┐\n│              三域并行启动（示例）                  │\n│                                                   │\n│  GEO域 → 市场研究(GEO-01) + 竞品分析(GEO-02)      │\n│        → 内容策略(GEO-03) + 多语言(GEO-04)        │\n│        → 平台适配(GEO-05) + 效果监测(GEO-06)     │\n│                                                   │\n│  亚马逊域 → 选品分析(AMZ-01) + Listing(AMZ-02)    │\n│           → 利润优化(AMZ-03) + 广告(AMZ-04)      │\n│           → 库存管理(AMZ-05) + 评价分析(AMZ-06)  │\n│                                                   │\n│  支撑域 → 数据采集(SUP-01) + 内容生成(SUP-02)     │\n│         → 翻译(SUP-03) + 合规检查(SUP-04)        │\n│         → 报告生成(SUP-05) + 质量评分(SUP-07)     │\n│         → 记忆管理(SUP-08) + 安全审计(SUP-10)    │\n└─────────────────────────────────────────────────┘\n    ↓\n结果聚合（幕僚长）\n    ↓\n报告生成(SUP-05)\n    ↓\n输出\n```\n\n### 3.2 Agent间P2P通信协议\n\n```python\n# Agent通信协议示例\nclass AgentMessage:\n    def __init__(self):\n        self.sender: str       # 发送方Agent ID\n        self.receiver: str     # 接收方Agent ID (空=广播)\n        self.msg_type: str     # REQUEST/RESPONSE/BROADCAST/EVENT\n        self.content: dict     # 消息内容\n        self.trace_id: str      # 全链路追踪ID\n        self.span_id: str      # 当前操作ID\n        self.parent_span_id: str # 父操作ID\n        self.priority: int    # 优先级 1-5\n        self.ttl: int          # 生存时间（秒）\n        self.timestamp: str   # ISO格式时间戳\n\n# 通信场景示例\n场景1: AMZ-01选品分析 → AMZ-02 Listing优化（新品上架）\n场景2: GEO-02竞品分析 → GEO-03内容策略（竞品差异化）\n场景3: SUP-01数据采集 → 所有消费者（数据供给）\n场景4: SUP-07质量评分 ← 所有生产者（质量监控）\n场景5: SUP-10安全审计 ← 所有Agent（安全事件上报）\n```\n\n### 3.3 依赖关系矩阵\n\n| 上游Agent | 下游Agent | 依赖类型 | 数据流向 |\n|---------|---------|---------|---------|\n| GEO-01 市场研究 | GEO-02/03, AMZ-01 | 强依赖 | 报告→策略 |\n| GEO-02 竞品分析 | GEO-03, AMZ-07 | 强依赖 | 画像→策略 |\n| GEO-03 内容策略 | GEO-04/05, SUP-02 | 强依赖 | 内容→多语言 |\n| GEO-06 效果监测 | 所有GEO Agent | 反馈循环 | 数据→优化 |\n| AMZ-01 选品 | AMZ-02/03/09 | 强依赖 | 选品→Listing |\n| AMZ-03 利润优化 | AMZ-04/08 | 强依赖 | 利润→定价/广告 |\n| AMZ-04 广告 | AMZ-09 | 中依赖 | 广告→关键词 |\n| AMZ-05 库存 | AMZ-08 | 中依赖 | 库存→定价 |\n| AMZ-06 评价 | AMZ-02 | 弱依赖 | 评价→Listing |\n| SUP-01 数据采集 | 所有消费者 | 强依赖 | 数据→各Agent |\n| SUP-07 质量评分 | 所有生产者 | 反馈 | 评分→改进 |\n| SUP-10 安全审计 | 所有Agent | 监控 | 审计→合规 |\n\n---\n\n## 四、Agent注册表\n\n### 4.1 完整注册表\n\n```json\n{\n  \"cluster_version\": \"3.0\",\n  \"cluster_name\": \"M-A3 30-Agent集群\",\n  \"chief_of_staff\": {\n    \"id\": \"chief-of-staff\",\n    \"name\": \"M-A3 幕僚长\",\n    \"domain\": \"orchestration\",\n    \"tier\": 0,\n    \"engine\": \"claude-ma\",\n    \"max_concurrent\": 1\n  },\n  \"agents\": [\n    /* ============ GEO域 ============ */\n    {\n      \"id\": \"geo-01-market-research\",\n      \"name\": \"市场研究Agent\",\n      \"domain\": \"geo\",\n      \"tier\": 2,\n      \"engine\": \"claude-ma\",\n      \"max_concurrent\": 3,\n      \"input_schema\": {\"type\": \"object\", \"required\": [\"product\", \"regions\"]},\n      \"output_schema\": {\"type\": \"object\", \"required\": [\"report\", \"opportunities\"]},\n      \"dependencies\": [],\n      \"upstream\": [\"chief-of-staff\"],\n      \"downstream\": [\"geo-02-competitor\", \"geo-03-content-strategy\", \"geo-10-regional\"],\n      \"keywords\": [\"市场研究\", \"市场规模\", \"目标市场\", \"TAM\", \"市场机会\"],\n      \"priority\": \"P0\",\n      \"skippable\": false,\n      \"timeout_seconds\": 300\n    },\n    {\n      \"id\": \"geo-02-competitor\",\n      \"name\": \"竞品分析Agent\",\n      \"domain\": \"geo\",\n      \"tier\": 2,\n      \"engine\": \"claude-ma\",\n      \"max_concurrent\": 3,\n      \"input_schema\": {\"type\": \"object\", \"required\": [\"competitors\", \"keywords\"]},\n      \"output_schema\": {\"type\": \"object\", \"required\": [\"geo_profile\", \"gap_analysis\"]},\n      \"dependencies\": [],\n      \"upstream\": [\"chief-of-staff\", \"geo-01-market-research\"],\n      \"downstream\": [\"geo-03-content-strategy\"],\n      \"keywords\": [\"竞品分析\", \"竞争对手\", \"对标分析\", \"竞品策略\"],\n      \"priority\": \"P1\",\n      \"skippable\": true,\n      \"timeout_seconds\": 240\n    },\n    {\n      \"id\": \"geo-03-content-strategy\",\n      \"name\": \"内容策略Agent\",\n      \"domain\": \"geo\",\n      \"tier\": 2,\n      \"engine\": \"claude-ma\",\n      \"max_concurrent\": 5,\n      \"input_schema\": {\"type\": \"object\", \"required\": [\"keywords\", \"product_info\"]},\n      \"output_schema\": {\"type\": \"object\", \"required\": [\"calendar\", \"topics\"]},\n      \"dependencies\": [\"geo-01-market-research\", \"geo-02-competitor\"],\n      \"upstream\": [\"chief-of-staff\"],\n      \"downstream\": [\"geo-04-multilingual\", \"geo-05-platform-adapt\", \"sup-02-content-gen\"],\n      \"keywords\": [\"内容策略\", \"内容规划\", \"选题\", \"内容日历\"],\n      \"priority\": \"P0\",\n      \"skippable\": false,\n      \"timeout_seconds\": 300\n    },\n    {\n      \"id\": \"geo-04-multilingual\",\n      \"name\": \"多语言优化Agent\",\n      \"domain\": \"geo\",\n      \"tier\": 2,\n      \"engine\": \"deepseek\",\n      \"max_concurrent\": 5,\n      \"input_schema\": {\"type\": \"object\", \"required\": [\"content\", \"languages\"]},\n      \"output_schema\": {\"type\": \"object\", \"required\": [\"localized_content\", \"hreflang\"]},\n      \"dependencies\": [\"geo-03-content-strategy\"],\n      \"upstream\": [\"geo-03-content-strategy\"],\n      \"downstream\": [\"sup-02-content-gen\", \"sup-03-translation\"],\n      \"keywords\": [\"多语言\", \"本地化\", \"翻译\", \"中译英\", \"国际化\"],\n      \"priority\": \"P1\",\n      \"skippable\": true,\n      \"timeout_seconds\": 180\n    },\n    {\n      \"id\": \"geo-05-platform-adapt\",\n      \"name\": \"平台适配Agent\",\n      \"domain\": \"geo\",\n      \"tier\": 2,\n      \"engine\": \"claude-ma\",\n      \"max_concurrent\": 4,\n      \"input_schema\": {\"type\": \"object\", \"required\": [\"content\", \"platforms\"]},\n      \"output_schema\": {\"type\": \"object\", \"required\": [\"adapted_content\", \"configs\"]},\n      \"dependencies\": [\"geo-03-content-strategy\"],\n      \"upstream\": [\"geo-03-content-strategy\"],\n      \"downstream\": [\"geo-06-monitoring\"],\n      \"keywords\": [\"平台适配\", \"多平台\", \"SEO适配\"],\n      \"priority\": \"P1\",\n      \"skippable\": true,\n      \"timeout_seconds\": 180\n    },\n    {\n      \"id\": \"geo-06-monitoring\",\n      \"name\": \"效果监测Agent\",\n      \"domain\": \"geo\",\n      \"tier\": 2,\n      \"engine\": \"local\",\n      \"max_concurrent\": 2,\n      \"input_schema\": {\"type\": \"object\", \"required\": [\"keywords\", \"date_range\"]},\n      \"output_schema\": {\"type\": \"object\", \"required\": [\"metrics\", \"report\"]},\n      \"dependencies\": [\"geo-05-platform-adapt\"],\n      \"upstream\": [\"geo-05-platform-adapt\", \"geo-03-content-strategy\"],\n      \"downstream\": [\"geo-02-competitor\", \"geo-03-content-strategy\"],\n      \"keywords\": [\"效果监测\", \"排名追踪\", \"AI搜索\", \"GEO效果\"],\n      \"priority\": \"P1\",\n      \"skippable\": true,\n      \"timeout_seconds\": 300\n    },\n    {\n      \"id\": \"geo-07-knowledge-graph\",\n      \"name\": \"知识图谱Agent\",\n      \"domain\": \"geo\",\n      \"tier\": 1,\n      \"engine\": \"claude-ma\",\n      \"max_concurrent\": 2,\n      \"input_schema\": {\"type\": \"object\", \"required\": [\"brand_info\", \"products\"]},\n      \"output_schema\": {\"type\": \"object\", \"required\": [\"knowledge_graph\", \"jsonld\"]},\n      \"dependencies\": [],\n      \"upstream\": [\"chief-of-staff\"],\n      \"downstream\": [\"geo-09-schema\", \"geo-03-content-strategy\"],\n      \"keywords\": [\"知识图谱\", \"Schema\", \"实体识别\", \"结构化数据\"],\n      \"priority\": \"P0\",\n      \"skippable\": false,\n      \"timeout_seconds\": 600\n    },\n    {\n      \"id\": \"geo-08-intent-prediction\",\n      \"name\": \"意图预测Agent\",\n      \"domain\": \"geo\",\n      \"tier\": 2,\n      \"engine\": \"claude-ma\",\n      \"max_concurrent\": 3,\n      \"input_schema\": {\"type\": \"object\", \"required\": [\"category\", \"keywords\"]},\n      \"output_schema\": {\"type\": \"object\", \"required\": [\"intent_matrix\", \"recommendations\"]},\n      \"dependencies\": [],\n      \"upstream\": [\"chief-of-staff\", \"geo-01-market-research\"],\n      \"downstream\": [\"geo-03-content-strategy\", \"amz-01-product-select\"],\n      \"keywords\": [\"意图预测\", \"搜索意图\", \"用户意图\"],\n      \"priority\": \"P1\",\n      \"skippable\": true,\n      \"timeout_seconds\": 240\n    },\n    {\n      \"id\": \"geo-09-schema\",\n      \"name\": \"Schema优化Agent\",\n      \"domain\": \"geo\",\n      \"tier\": 2,\n      \"engine\": \"local\",\n      \"max_concurrent\": 4,\n      \"input_schema\": {\"type\": \"object\", \"required\": [\"urls\"]},\n      \"output_schema\": {\"type\": \"object\", \"required\": [\"audit_report\", \"jsonld_snippets\"]},\n      \"dependencies\": [\"geo-07-knowledge-graph\"],\n      \"upstream\": [\"geo-07-knowledge-graph\"],\n      \"downstream\": [],\n      \"keywords\": [\"Schema优化\", \"结构化数据\", \"Rich Results\"],\n      \"priority\": \"P2\",\n      \"skippable\": true,\n      \"timeout_seconds\": 300\n    },\n    {\n      \"id\": \"geo-10-regional\",\n      \"name\": \"地域策略Agent\",\n      \"domain\": \"geo\",\n      \"tier\": 2,\n      \"engine\": \"claude-ma\",\n      \"max_concurrent\": 3,\n      \"input_schema\": {\"type\": \"object\", \"required\": [\"brand\", \"regions\"]},\n      \"output_schema\": {\"type\": \"object\", \"required\": [\"regional_strategies\"]},\n      \"dependencies\": [\"geo-01-market-research\"],\n      \"upstream\": [\"geo-01-market-research\"],\n      \"downstream\": [\"geo-03-content-strategy\"],\n      \"keywords\": [\"地域策略\", \"本地SEO\", \"ccTLD\"],\n      \"priority\": \"P2\",\n      \"skippable\": true,\n      \"timeout_seconds\": 240\n    },\n\n    /* ============ 亚马逊域 ============ */\n    {\n      \"id\": \"amz-01-product-select\",\n      \"name\": \"选品分析Agent\",\n      \"domain\": \"amazon\",\n      \"tier\": 2,\n      \"engine\": \"claude-ma\",\n      \"max_concurrent\": 3,\n      \"input_schema\": {\"type\": \"object\", \"required\": [\"category\", \"budget\"]},\n      \"output_schema\": {\"type\": \"object\", \"required\": [\"top_products\", \"risk_ratings\"]},\n      \"dependencies\": [],\n      \"upstream\": [\"chief-of-staff\", \"geo-01-market-research\"],\n      \"downstream\": [\"amz-02-listing\", \"amz-03-profit\", \"amz-09-keywords\"],\n      \"keywords\": [\"选品\", \"新品开发\", \"产品机会\"],\n      \"priority\": \"P0\",\n      \"skippable\": false,\n      \"timeout_seconds\": 300\n    },\n    {\n      \"id\": \"amz-02-listing\",\n      \"name\": \"Listing优化Agent\",\n      \"domain\": \"amazon\",\n      \"tier\": 2,\n      \"engine\": \"claude-ma\",\n      \"max_concurrent\": 5,\n      \"input_schema\": {\"type\": \"object\", \"required\": [\"product_info\", \"competitor_asins\"]},\n      \"output_schema\": {\"type\": \"object\", \"required\": [\"title\", \"bullets\", \"description\", \"score\"]},\n      \"dependencies\": [\"amz-01-product-select\"],\n      \"upstream\": [\"amz-01-product-select\"],\n      \"downstream\": [\"amz-04-ads\", \"sup-04-compliance\"],\n      \"keywords\": [\"Listing优化\", \"标题优化\", \"五点描述\", \"产品页面\"],\n      \"priority\": \"P0\",\n      \"skippable\": false,\n      \"timeout_seconds\": 300\n    },\n    {\n      \"id\": \"amz-03-profit\",\n      \"name\": \"利润优化Agent\",\n      \"domain\": \"amazon\",\n      \"tier\": 2,\n      \"engine\": \"claude-ma\",\n      \"max_concurrent\": 3,\n      \"input_schema\": {\"type\": \"object\", \"required\": [\"asins\", \"cost_structure\"]},\n      \"output_schema\": {\"type\": \"object\", \"required\": [\"profit_report\", \"pricing_suggestions\"]},\n      \"dependencies\": [\"amz-01-product-select\"],\n      \"upstream\": [\"amz-01-product-select\"],\n      \"downstream\": [\"amz-04-ads\", \"amz-08-pricing\"],\n      \"keywords\": [\"利润优化\", \"ACOS\", \"TACOS\", \"成本分析\"],\n      \"priority\": \"P0\",\n      \"skippable\": false,\n      \"timeout_seconds\": 300\n    },\n    {\n      \"id\": \"amz-04-ads\",\n      \"name\": \"广告投放Agent\",\n      \"domain\": \"amazon\",\n      \"tier\": 2,\n      \"engine\": \"claude-ma\",\n      \"max_concurrent\": 4,\n      \"input_schema\": {\"type\": \"object\", \"required\": [\"asin\", \"budget\", \"target_acos\"]},\n      \"output_schema\": {\"type\": \"object\", \"required\": [\"campaign_structure\", \"keyword_bids\"]},\n      \"dependencies\": [\"amz-02-listing\", \"amz-03-profit\"],\n      \"upstream\": [\"amz-02-listing\", \"amz-03-profit\"],\n      \"downstream\": [\"amz-09-keywords\"],\n      \"keywords\": [\"广告投放\", \"SP广告\", \"SB广告\", \"ACOS优化\"],\n      \"priority\": \"P0\",\n      \"skippable\": false,\n      \"timeout_seconds\": 300\n    },\n    {\n      \"id\": \"amz-05-inventory\",\n      \"name\": \"库存管理Agent\",\n      \"domain\": \"amazon\",\n      \"tier\": 2,\n      \"engine\": \"local\",\n      \"max_concurrent\": 3,\n      \"input_schema\": {\"type\": \"object\", \"required\": [\"asins\", \"inventory_data\"]},\n      \"output_schema\": {\"type\": \"object\", \"required\": [\"replenishment_plan\", \"alerts\"]},\n      \"dependencies\": [],\n      \"upstream\": [\"chief-of-staff\"],\n      \"downstream\": [\"amz-08-pricing\"],\n      \"keywords\": [\"库存管理\", \"补货\", \"FBA库存\"],\n      \"priority\": \"P1\",\n      \"skippable\": false,\n      \"timeout_seconds\": 180\n    },\n    {\n      \"id\": \"amz-06-review\",\n      \"name\": \"评价分析Agent\",\n      \"domain\": \"amazon\",\n      \"tier\": 2,\n      \"engine\": \"claude-ma\",\n      \"max_concurrent\": 4,\n      \"input_schema\": {\"type\": \"object\", \"required\": [\"asins\"]},\n      \"output_schema\": {\"type\": \"object\", \"required\": [\"review_report\", \"action_items\"]},\n      \"dependencies\": [],\n      \"upstream\": [\"chief-of-staff\", \"amz-02-listing\"],\n      \"downstream\": [\"amz-02-listing\", \"amz-06-review\"],\n      \"keywords\": [\"评价分析\", \"review\", \"QA分析\", \"差评监控\"],\n      \"priority\": \"P1\",\n      \"skippable\": true,\n      \"timeout_seconds\": 240\n    },\n    {\n      \"id\": \"amz-07-competitor-monitor\",\n      \"name\": \"竞品监控Agent\",\n      \"domain\": \"amazon\",\n      \"tier\": 2,\n      \"engine\": \"local\",\n      \"max_concurrent\": 3,\n      \"input_schema\": {\"type\": \"object\", \"required\": [\"competitor_asins\"]},\n      \"output_schema\": {\"type\": \"object\", \"required\": [\"daily_report\", \"alerts\"]},\n      \"dependencies\": [],\n      \"upstream\": [\"chief-of-staff\"],\n      \"downstream\": [\"amz-08-pricing\", \"amz-04-ads\"],\n      \"keywords\": [\"竞品监控\", \"价格战\", \"市场情报\"],\n      \"priority\": \"P1\",\n      \"skippable\": true,\n      \"timeout_seconds\": 300\n    },\n    {\n      \"id\": \"amz-08-pricing\",\n      \"name\": \"定价策略Agent\",\n      \"domain\": \"amazon\",\n      \"tier\": 2,\n      \"engine\": \"claude-ma\",\n      \"max_concurrent\": 3,\n      \"input_schema\": {\"type\": \"object\", \"required\": [\"asin\", \"cost\", \"competition\"]},\n      \"output_schema\": {\"type\": \"object\", \"required\": [\"pricing_strategy\", \"rules\"]},\n      \"dependencies\": [\"amz-03-profit\", \"amz-05-inventory\", \"amz-07-competitor-monitor\"],\n      \"upstream\": [\"amz-03-profit\", \"amz-05-inventory\", \"amz-07-competitor-monitor\"],\n      \"downstream\": [],\n      \"keywords\": [\"定价策略\", \"动态定价\", \"Buy Box\"],\n      \"priority\": \"P1\",\n      \"skippable\": true,\n      \"timeout_seconds\": 240\n    },\n    {\n      \"id\": \"amz-09-keywords\",\n      \"name\": \"关键词Agent\",\n      \"domain\": \"amazon\",\n      \"tier\": 1,\n      \"engine\": \"claude-ma\",\n      \"max_concurrent\": 3,\n      \"input_schema\": {\"type\": \"object\", \"required\": [\"asin\", \"seed_keywords\"]},\n      \"output_schema\": {\"type\": \"object\", \"required\": [\"keyword_matrix\", \"priorities\"]},\n      \"dependencies\": [\"amz-01-product-select\"],\n      \"upstream\": [\"amz-01-product-select\", \"amz-04-ads\"],\n      \"downstream\": [\"amz-02-listing\", \"amz-04-ads\"],\n      \"keywords\": [\"关键词\", \"keyword\", \"长尾词\", \"否定词\"],\n      \"priority\": \"P1\",\n      \"skippable\": false,\n      \"timeout_seconds\": 300\n    },\n    {\n      \"id\": \"amz-10-reporting\",\n      \"name\": \"报表分析Agent\",\n      \"domain\": \"amazon\",\n      \"tier\": 2,\n      \"engine\": \"local\",\n      \"max_concurrent\": 2,\n      \"input_schema\": {\"type\": \"object\", \"required\": [\"report_type\", \"date_range\"]},\n      \"output_schema\": {\"type\": \"object\", \"required\": [\"report_file\", \"summary\"]},\n      \"dependencies\": [],\n      \"upstream\": [\"chief-of-staff\"],\n      \"downstream\": [],\n      \"keywords\": [\"报表\", \"KPI\", \"周报\", \"月报\", \"数据分析\"],\n      \"priority\": \"P2\",\n      \"skippable\": true,\n      \"timeout_seconds\": 180\n    },\n\n    /* ============ 支撑域 ============ */\n    {\n      \"id\": \"sup-01-data-collect\",\n      \"name\": \"数据采集Agent\",\n      \"domain\": \"support\",\n      \"tier\": 1,\n      \"engine\": \"local\",\n      \"max_concurrent\": 5,\n      \"input_schema\": {\"type\": \"object\", \"required\": [\"sources\", \"data_requirements\"]},\n      \"output_schema\": {\"type\": \"object\", \"required\": [\"dataset\", \"quality_report\"]},\n      \"dependencies\": [],\n      \"upstream\": [\"chief-of-staff\"],\n      \"downstream\": [\"geo-01-market-research\", \"amz-01-product-select\", \"sup-07-quality\"],\n      \"keywords\": [\"数据采集\", \"爬虫\", \"数据清洗\"],\n      \"priority\": \"P0\",\n      \"skippable\": false,\n      \"timeout_seconds\": 600\n    },\n    {\n      \"id\": \"sup-02-content-gen\",\n      \"name\": \"内容生成Agent\",\n      \"domain\": \"support\",\n      \"tier\": 1,\n      \"engine\": \"claude-ma\",\n      \"max_concurrent\": 8,\n      \"input_schema\": {\"type\": \"object\", \"required\": [\"content_type\", \"product_info\"]},\n      \"output_schema\": {\"type\": \"object\", \"required\": [\"generated_content\", \"quality_score\"]},\n      \"dependencies\": [\"geo-03-content-strategy\"],\n      \"upstream\": [\"geo-03-content-strategy\"],\n      \"downstream\": [\"sup-04-compliance\", \"sup-07-quality\"],\n      \"keywords\": [\"内容生成\", \"批量文案\", \"脚本生成\"],\n      \"priority\": \"P0\",\n      \"skippable\": false,\n      \"timeout_seconds\": 300\n    },\n    {\n      \"id\": \"sup-03-translation\",\n      \"name\": \"翻译Agent\",\n      \"domain\": \"support\",\n      \"tier\": 1,\n      \"engine\": \"deepseek\",\n      \"max_concurrent\": 10,\n      \"input_schema\": {\"type\": \"object\", \"required\": [\"text\", \"target_language\", \"industry\"]},\n      \"output_schema\": {\"type\": \"object\", \"required\": [\"translated_text\", \"quality_score\"]},\n      \"dependencies\": [],\n      \"upstream\": [\"geo-04-multilingual\", \"amz-02-listing\"],\n      \"downstream\": [\"sup-07-quality\"],\n      \"keywords\": [\"翻译\", \"中英翻译\", \"本地化\"],\n      \"priority\": \"P1\",\n      \"skippable\": true,\n      \"timeout_seconds\": 120\n    },\n    {\n      \"id\": \"sup-04-compliance\",\n      \"name\": \"合规检查Agent\",\n      \"domain\": \"support\",\n      \"tier\": 1,\n      \"engine\": \"deepseek\",\n      \"max_concurrent\": 5,\n      \"input_schema\": {\"type\": \"object\", \"required\": [\"content\", \"platform\", \"market\"]},\n      \"output_schema\": {\"type\": \"object\", \"required\": [\"compliance_report\", \"risk_rating\"]},\n      \"dependencies\": [],\n      \"upstream\": [\"amz-02-listing\", \"sup-02-content-gen\"],\n      \"downstream\": [\"sup-07-quality\"],\n      \"keywords\": [\"合规检查\", \"政策合规\", \"风险审核\"],\n      \"priority\": \"P0\",\n      \"skippable\": false,\n      \"timeout_seconds\": 180\n    },\n    {\n      \"id\": \"sup-05-report-gen\",\n      \"name\": \"报告生成Agent\",\n      \"domain\": \"support\",\n      \"tier\": 2,\n      \"engine\": \"claude-ma\",\n      \"max_concurrent\": 3,\n      \"input_schema\": {\"type\": \"object\", \"required\": [\"report_type\", \"data\"]},\n      \"output_schema\": {\"type\": \"object\", \"required\": [\"report_file\", \"summary\"]},\n      \"dependencies\": [],\n      \"upstream\": [\"chief-of-staff\"],\n      \"downstream\": [],\n      \"keywords\": [\"报告生成\", \"PPT\", \"数据分析报告\"],\n      \"priority\": \"P1\",\n      \"skippable\": false,\n      \"timeout_seconds\": 300\n    },\n    {\n      \"id\": \"sup-06-customer-service\",\n      \"name\": \"客户服务Agent\",\n      \"domain\": \"support\",\n      \"tier\": 2,\n      \"engine\": \"claude-ma\",\n      \"max_concurrent\": 10,\n      \"input_schema\": {\"type\": \"object\", \"required\": [\"query\", \"history\"]},\n      \"output_schema\": {\"type\": \"object\", \"required\": [\"response\", \"action\"]},\n      \"dependencies\": [],\n      \"upstream\": [\"chief-of-staff\"],\n      \"downstream\": [\"sup-07-quality\"],\n      \"keywords\": [\"客服\", \"自动回复\", \"FAQ\"],\n      \"priority\": \"P2\",\n      \"skippable\": true,\n      \"timeout_seconds\": 60\n    },\n    {\n      \"id\": \"sup-07-quality\",\n      \"name\": \"质量评分Agent\",\n      \"domain\": \"support\",\n      \"tier\": 1,\n      \"engine\": \"claude-ma\",\n      \"max_concurrent\": 5,\n      \"input_schema\": {\"type\": \"object\", \"required\": [\"output\", \"task_type\"]},\n      \"output_schema\": {\"type\": \"object\", \"required\": [\"quality_score\", \"improvements\"]},\n      \"dependencies\": [],\n      \"upstream\": [\"sup-01-data-collect\", \"sup-02-content-gen\", \"sup-03-translation\", \"sup-06-customer-service\"],\n      \"downstream\": [\"chief-of-staff\"],\n      \"keywords\": [\"质量评分\", \"效果评估\"],\n      \"priority\": \"P1\",\n      \"skippable\": true,\n      \"timeout_seconds\": 120\n    },\n    {\n      \"id\": \"sup-08-memory\",\n      \"name\": \"记忆管理Agent\",\n      \"domain\": \"support\",\n      \"tier\": 1,\n      \"engine\": \"local\",\n      \"max_concurrent\": 20,\n      \"input_schema\": {\"type\": \"object\", \"required\": [\"context\", \"operation\"]},\n      \"output_schema\": {\"type\": \"object\", \"required\": [\"memories\", \"updated_context\"]},\n      \"dependencies\": [],\n      \"upstream\": [\"chief-of-staff\"],\n      \"downstream\": [\"chief-of-staff\"],\n      \"keywords\": [\"记忆管理\", \"知识库\", \"上下文\"],\n      \"priority\": \"P1\",\n      \"skippable\": false,\n      \"timeout_seconds\": 60\n    },\n    {\n      \"id\": \"sup-09-scheduler\",\n      \"name\": \"调度协调Agent\",\n      \"domain\": \"support\",\n      \"tier\": 0,\n      \"engine\": \"local\",\n      \"max_concurrent\": 1,\n      \"input_schema\": {\"type\": \"object\", \"required\": [\"tasks\", \"constraints\"]},\n      \"output_schema\": {\"type\": \"object\", \"required\": [\"execution_plan\", \"schedule_log\"]},\n      \"dependencies\": [],\n      \"upstream\": [\"chief-of-staff\"],\n      \"downstream\": [\"chief-of-staff\"],\n      \"keywords\": [\"任务调度\", \"并行执行\", \"工作流\"],\n      \"priority\": \"P0\",\n      \"skippable\": false,\n      \"timeout_seconds\": 120\n    },\n    {\n      \"id\": \"sup-10-security\",\n      \"name\": \"安全审计Agent\",\n      \"domain\": \"support\",\n      \"tier\": 0,\n      \"engine\": \"local\",\n      \"max_concurrent\": 2,\n      \"input_schema\": {\"type\": \"object\", \"required\": [\"scope\", \"agents\"]},\n      \"output_schema\": {\"type\": \"object\", \"required\": [\"audit_report\", \"risk_items\"]},\n      \"dependencies\": [],\n      \"upstream\": [\"chief-of-staff\"],\n      \"downstream\": [\"chief-of-staff\"],\n      \"keywords\": [\"安全审计\", \"权限管理\", \"日志审计\"],\n      \"priority\": \"P0\",\n      \"skippable\": false,\n      \"timeout_seconds\": 300\n    }\n  ],\n\n  \"domain_summary\": {\n    \"geo\": {\"count\": 10, \"p0\": 4, \"p1\": 4, \"p2\": 2},\n    \"amazon\": {\"count\": 10, \"p0\": 4, \"p1\": 5, \"p2\": 1},\n    \"support\": {\"count\": 10, \"p0\": 5, \"p1\": 4, \"p2\": 1}\n  },\n\n  \"engine_distribution\": {\n    \"claude-ma\": 18,\n    \"local\": 8,\n    \"deepseek\": 4\n  }\n}\n```\n\n---\n\n## 五、资源管理与扩展策略\n\n### 5.1 Agent池配置\n\n```yaml\n# agent-pool-config.yaml\nagent_pool:\n  max_total_agents: 30\n  max_concurrent_tasks: 50\n  idle_timeout_seconds: 300\n  \n  domains:\n    geo:\n      agents: 10\n      max_concurrent_per_agent: 5\n      shared_context_size_mb: 512\n      \n    amazon:\n      agents: 10\n      max_concurrent_per_agent: 5\n      shared_context_size_mb: 512\n      \n    support:\n      agents: 10\n      max_concurrent_per_agent: 10\n      shared_context_size_mb: 256\n\n  engines:\n    claude-ma:\n      max_instances: 18\n      token_limit: 200000\n      fallback: \"local\"\n      \n    local:\n      max_instances: 8\n      token_limit: 32000\n      fallback: \"deepseek\"\n      \n    deepseek:\n      max_instances: 4\n      token_limit: 64000\n      fallback: \"local\"\n\n  rate_limits:\n    per_agent_per_minute: 60\n    per_domain_per_minute: 200\n    cluster_wide_per_minute: 500\n```\n\n### 5.2 负载均衡策略\n\n```\n┌─────────────────────────────────────────────┐\n│           幕僚长请求分发算法                    │\n├─────────────────────────────────────────────┤\n│  1. 意图分类 → 功能域（geo/amazon/support）   │\n│  2. 功能域内 → 关键词匹配 → Agent ID          │\n│  3. Agent选择 → 负载均衡（Least Connections） │\n│  4. 依赖检查 → DAG排序 → 并行组              │\n│  5. 资源预检 → 容量检查 → 执行/排队/拒绝      │\n└─────────────────────────────────────────────┘\n\n负载均衡算法：加权最少连接（Weighted Least Connections）\n- 权重 = Agent最大并发数\n- 活跃连接数 = 当前任务数\n- 选择：权重 - 活跃连接数 最大的Agent\n```\n\n### 5.3 降级策略\n\n| 触发条件 | 降级动作 |\n|---------|---------|\n| Engine不可用 | 切换备用引擎（Claude→Local→DeepSeek） |\n| Agent过载 | 任务进入优先级队列，等待释放 |\n| Token超限 | 压缩上下文，保留关键记忆 |\n| 网络故障 | 缓存结果，降级为只读模式 |\n| 安全事件 | 隔离受影响Agent，启动审计 |\n\n---\n\n## 六、质量保障体系\n\n### 6.1 质量评分维度\n\n```python\nQUALITY_DIMENSIONS = {\n    \"accuracy\":      {\"weight\": 0.30, \"desc\": \"信息准确性\"},\n    \"completeness\":  {\"weight\": 0.25, \"desc\": \"任务完成度\"},\n    \"actionability\": {\"weight\": 0.20, \"desc\": \"建议可执行性\"},\n    \"timeliness\":    {\"weight\": 0.15, \"desc\": \"时效性\"},\n    \"clarity\":       {\"weight\": 0.10, \"desc\": \"表达清晰度\"},\n}\n\n# 评分等级\nEXCELLENT = (90, 100)  # 无需审查\nGOOD      = (75, 89)   # 可选审查\nFAIR      = (60, 74)   # 建议审查\nPOOR      = (0, 59)    # 必须审查\n```\n\n### 6.2 全链路追踪\n\n```\nTrace-ID: geo-20260414-001\n├─ span: chief-of-staff.intent-recognition (50ms)\n├─ span: sup-09-scheduler.task-decomposition (120ms)\n└─ span: parallel-execution-group\n   ├─ [并行] geo-01-market-research (280s) ✓\n   ├─ [并行] geo-02-competitor-analysis (195s) ✓\n   ├─ [并行] amz-01-product-select (310s) ✓\n   ├─ [并行] sup-01-data-collection (180s) ✓\n   └─ [串行] geo-03-content-strategy (等待geo-01/02完成)\n       ├─ [并行] geo-04-multilingual\n       └─ [并行] geo-05-platform-adapt\n```\n\n---\n\n## 七、竞争优势分析\n\n### 7.1 竞品对比\n\n| 维度 | 普通竞品 | 本方案 |\n|------|---------|--------|\n| Agent数量 | 3-8个 | **30个** |\n| 功能域 | 单一域 | **三域全覆盖** |\n| 协作深度 | 串行为主 | **并行+串行混合** |\n| 商业闭环 | 部分 | **全链路（GEO→Amazon→运营）** |\n| 自进化能力 | 无 | **记忆+学习+改进** |\n| 质量保障 | 无 | **SUP-07质量评分** |\n| 安全体系 | 基础 | **SUP-10安全审计+MCP安全扫描** |\n\n### 7.2 核心差异化壁垒\n\n1. **规模壁垒**：30个垂直Agent覆盖全链路，竞品难以快速复制\n2. **数据壁垒**：跨域数据流动（Geo-Amazon数据互通）形成网络效应\n3. **自进化壁垒**：记忆管理Agent持续积累行业知识，时间越久越强\n4. **质量壁垒**：SUP-07质量评分体系确保每项输出可量化、可改进\n5. **安全壁垒**：SUP-10安全审计主动防护MCP协议漏洞（43%市场风险）\n\n---\n\n*本方案由 M-A3 幕僚长 设计，版本 v1.0*\n*生成时间：2026-04-14*\n\nFile v3.0.4:30-expansion/agent_protocol.md\n\n# M-A3 30-Agent集群 通信协议规范\n> 版本：v1.0 | 日期：2026-04-14\n\n---\n\n## 一、协议概述\n\n### 1.1 设计目标\n- **可靠性**：消息可靠投递，支持重试和确认\n- **可追踪性**：全链路 trace_id，支持逆向追溯\n- **高效性**：支持并行/串行动态切换\n- **安全性**：消息加密 + 权限校验\n\n### 1.2 协议层次\n\n```\n┌─────────────────────────────────────────────────────┐\n│              Agent Message Protocol (AMP)            │\n├─────────────────────────────────────────────────────┤\n│  Layer 4: 业务语义层  (Task/Broadcast/Event/Ack)     │\n│  Layer 3: 路由层     (P2P/Broadcast/Fan-out)         │\n│  Layer 2: 可靠性层   (Retry/Ack/Timeout/Dedup)      │\n│  Layer 1: 传输层     (HTTP-LongPoll/WebSocket/gRPC) │\n└─────────────────────────────────────────────────────┘\n```\n\n---\n\n## 二、消息格式\n\n### 2.1 AgentMessage 结构\n\n```python\nfrom dataclasses import dataclass, field\nfrom typing import Optional, Any\nfrom datetime import datetime\n\n@dataclass\nclass AgentMessage:\n    # ── 身份标识 ──────────────────────────────────────\n    msg_id: str                         # 全局唯一消息ID (UUIDv7)\n    trace_id: str                       # 全链路追踪ID\n    span_id: str                        # 当前操作ID\n    parent_span_id: Optional[str]       # 父操作ID (顶级无parent)\n    \n    # ── 通信元数据 ─────────────────────────────────────\n    sender: str                         # 发送方Agent ID\n    receiver: str                       # 接收方Agent ID (\"*\" = 广播)\n    msg_type: str                       # REQUEST/RESPONSE/BROADCAST/EVENT/ACK\n    priority: int                       # 1-5 (1最高)\n    \n    # ── 业务载荷 ───────────────────────────────────────\n    action: str                         # 操作类型 (见操作清单)\n    payload: dict                       # 消息内容\n    expected_response_format: Optional[dict] = None  # 响应格式约定\n    \n    # ── 可靠性 ─────────────────────────────────────────\n    ttl_seconds: int = 300              # 生存时间\n    retry_count: int = 0                # 已重试次数\n    max_retries: int = 3                # 最大重试次数\n    correlation_id: Optional[str] = None # 关联请求ID (用于响应匹配)\n    \n    # ── 安全 ───────────────────────────────────────────\n    auth_token: Optional[str] = None\n    pii_present: bool = False           # 是否含敏感信息\n    \n    # ── 时间戳 ─────────────────────────────────────────\n    created_at: str = field(default_factory=lambda: datetime.utcnow().isoformat())\n    expires_at: Optional[str] = None\n\n# ── 消息类型枚举 ──────────────────────────────────────\nclass MsgType:\n    REQUEST   = \"REQUEST\"    # 请求-响应模式\n    RESPONSE  = \"RESPONSE\"   # 响应消息\n    BROADCAST = \"BROADCAST\"  # 广播（无响应期望）\n    EVENT     = \"EVENT\"      # 事件通知（异步）\n    ACK       = \"ACK\"        # 确认消息\n\n# ── 操作类型清单 ──────────────────────────────────────\nclass AgentAction:\n    # 通用\n    HEALTH_CHECK    = \"health_check\"\n    REPORT_STATUS   = \"report_status\"\n    FETCH_DATA      = \"fetch_data\"\n    \n    # GEO域\n    MARKET_RESEARCH = \"market_research\"\n    COMPETITOR_ANALYSIS = \"competitor_analysis\"\n    CONTENT_STRATEGY = \"content_strategy\"\n    MULTILINGUAL_OPT = \"multilingual_optimization\"\n    PLATFORM_ADAPT   = \"platform_adaptation\"\n    MONITORING       = \"geo_monitoring\"\n    KNOWLEDGE_GRAPH  = \"knowledge_graph_build\"\n    INTENT_PREDICT   = \"intent_prediction\"\n    SCHEMA_OPT       = \"schema_optimization\"\n    REGIONAL_STRATEGY = \"regional_strategy\"\n    \n    # 亚马逊域\n    PRODUCT_SELECT   = \"product_selection\"\n    LISTING_OPT      = \"listing_optimization\"\n    PROFIT_OPT       = \"profit_optimization\"\n    ADS_MANAGE       = \"ads_management\"\n    INVENTORY_MANAGE = \"inventory_management\"\n    REVIEW_ANALYSIS  = \"review_analysis\"\n    COMPETITOR_MONITOR = \"competitor_monitoring\"\n    PRICING_STRATEGY = \"pricing_strategy\"\n    KEYWORD_RESEARCH = \"keyword_research\"\n    REPORTING        = \"report_generation\"\n    \n    # 支撑域\n    DATA_COLLECT     = \"data_collection\"\n    CONTENT_GENERATE  = \"content_generation\"\n    TRANSLATE         = \"translation\"\n    COMPLIANCE_CHECK  = \"compliance_check\"\n    REPORT_GENERATE   = \"report_generate\"\n    CUSTOMER_SERVICE  = \"customer_service\"\n    QUALITY_SCORING   = \"quality_scoring\"\n    MEMORY_MANAGE     = \"memory_management\"\n    SCHEDULE_TASK     = \"schedule_task\"\n    SECURITY_AUDIT    = \"security_audit\"\n```\n\n### 2.2 响应格式\n\n```python\n@dataclass\nclass AgentResponse:\n    msg_id: str                        # 对应请求的 msg_id\n    trace_id: str                      # 继承请求的 trace_id\n    span_id: str                       # 响应操作的 span_id\n    \n    status: str                        # SUCCESS / PARTIAL / FAILED / TIMEOUT\n    error_code: Optional[str] = None   # 错误码\n    error_message: Optional[str] = None\n    \n    result: Optional[Any] = None       # 业务结果\n    quality_score: Optional[float] = None  # SUP-07评分\n    \n    metadata: dict = field(default_factory=dict)\n    # metadata 可包含: execution_time_ms, tokens_used, agent_id\n    \n    created_at: str = field(default_factory=lambda: datetime.utcnow().isoformat())\n```\n\n---\n\n## 三、路由模式\n\n### 3.1 路由类型\n\n| 模式 | 说明 | 使用场景 |\n|------|------|---------|\n| `P2P` | 点对点，一对一 | 明确的下游Agent |\n| `FAN_IN` | 多对一，聚合 | 结果汇总 |\n| `FAN_OUT` | 一对多，并行 | 多域并行启动 |\n| `BROADCAST` | 广播至所有Agent | 系统公告/安全事件 |\n| `DOMAIN_BROADCAST` | 域内广播 | GEO域内通知 |\n\n### 3.2 路由配置示例\n\n```yaml\n# 典型任务路由配置\nrouting_templates:\n  # 场景1：新品上市全链路（GEO + 亚马逊并行）\n  new_product_launch:\n    parallel_groups:\n      - domain: \"geo\"\n        agents: [\"geo-01-market-research\", \"geo-02-competitor\"]\n        mode: \"PARALLEL\"\n      - domain: \"amazon\"\n        agents: [\"amz-01-product-select\"]\n        mode: \"PARALLEL\"\n    dependencies:\n      - from: \"geo-01-market-research\"\n        to: \"geo-03-content-strategy\"\n      - from: \"geo-02-competitor\"\n        to: \"geo-03-content-strategy\"\n      - from: [\"geo-01-market-research\", \"amz-01-product-select\"]\n        to: \"sup-09-scheduler\"\n        mode: \"FAN_IN\"\n\n  # 场景2：竞品动态监控\n  competitor_monitoring:\n    agents: [\"amz-07-competitor-monitor\"]\n    schedule: \"cron:0 * * * *\"  # 每小时\n    downstream:\n      - agent: \"amz-08-pricing\"\n        condition: \"price_change > 5%\"\n      - agent: \"amz-04-ads\"\n        condition: \"competitor_rank_change\"\n\n  # 场景3：质量问题升级\n  quality_escalation:\n    trigger: \"quality_score < 60\"\n    agents: [\"sup-07-quality\", \"sup-10-security\"]\n    notify: \"chief-of-staff\"\n    severity: \"high\"\n```\n\n---\n\n## 四、可靠性机制\n\n### 4.1 重试策略\n\n```python\nRETRY_CONFIG = {\n    \"max_retries\": 3,\n    \"backoff\": {\n        \"type\": \"exponential\",\n        \"initial_ms\": 1000,\n        \"multiplier\": 2.0,\n        \"max_ms\": 30000,\n        \"jitter\": True  # ±10%随机抖动避免惊群\n    },\n    \"retryable_errors\": [\n        \"NETWORK_ERROR\",\n        \"TIMEOUT\",\n        \"SERVICE_UNAVAILABLE\",\n        \"RATE_LIMITED\"\n    ],\n    \"non_retryable_errors\": [\n        \"INVALID_REQUEST\",\n        \"UNAUTHORIZED\",\n        \"FORBIDDEN\",\n        \"NOT_FOUND\"\n    ]\n}\n```\n\n### 4.2 幂等性保证\n\n```python\n# 每个 REQUEST 消息携带 idempotency_key\n# 接收方基于 (sender, action, idempotency_key) 做去重\n# 相同key的重复请求直接返回缓存响应\n\n@dataclass\nclass IdempotencyRecord:\n    key: str              # hash(msg_id)\n    request_hash: str     # hash(payload)\n    response: AgentResponse\n    created_at: datetime\n    expires_at: datetime  # TTL=3600s\n```\n\n### 4.3 超时配置\n\n```python\nTIMEOUT_CONFIG = {\n    # P2P 请求超时（按操作类型）\n    \"P2P_DEFAULT\": 30,          # 秒\n    \"market_research\": 300,\n    \"knowledge_graph_build\": 600,\n    \"data_collection\": 600,\n    \"content_generation\": 300,\n    \"listing_optimization\": 300,\n    \"translation\": 120,\n    \"health_check\": 5,\n    \"memory_management\": 60,\n    \"quality_scoring\": 120,\n    \n    # 整链路超时（按场景）\n    \"GEO_FULL_PIPELINE\": 1800,   # 30分钟\n    \"AMAZON_FULL_PIPELINE\": 1800,\n    \"NEW_PRODUCT_LAUNCH\": 3600   # 60分钟\n}\n```\n\n---\n\n## 五、安全协议\n\n### 5.1 消息加密\n\n```python\n# 所有跨域消息必须加密\nMESSAGE_SECURITY = {\n    \"encryption\": \"AES-256-GCM\",      # 消息体加密\n    \"signature\": \"HMAC-SHA256\",       # 完整性校验\n    \"key_exchange\": \"ECDH-P256\",      # 密钥协商\n    \n    # Token格式\n    \"token_format\": \"Bearer {jwt}\",\n    \"jwt_algorithm\": \"RS256\",\n    \"token_expiry_seconds\": 3600,\n    \n    # 敏感字段\n    \"pii_fields\": [\"email\", \"phone\", \"id_card\", \"bank_account\", \"api_key\"],\n    \"pii_action\": \"MASK\"              # MASK/REJECT/LOG\n}\n```\n\n### 5.2 权限矩阵\n\n```python\n# Agent间调用权限 (简化示例)\nPERMISSION_MATRIX = {\n    # 格式: (caller, action, resource) -> allowed\n    # 各Agent调用基础设施Agent\n    (\"*\", \"health_check\", \"sup-10-security\"): True,\n    (\"*\", \"security_audit\", \"sup-10-security\"): True,\n    (\"*\", \"fetch_data\", \"sup-01-data-collect\"): True,\n    (\"*\", \"memory_management\", \"sup-08-memory\"): True,\n    \n    # 数据流权限\n    (\"geo-01-market-research\", \"REQUEST\", \"sup-01-data-collect\"): True,\n    (\"amz-01-product-select\", \"REQUEST\", \"sup-01-data-collect\"): True,\n    \n    # 质量监控\n    (\"*\", \"quality_scoring\", \"sup-07-quality\"): True,\n    \n    # 跨域数据流\n    (\"geo-01-market-research\", \"REQUEST\", \"amz-01-product-select\"): True,  # GEO→Amazon数据共享\n    (\"amz-06-review\", \"FETCH_DATA\", \"geo-07-knowledge-graph\"): True,      # Amazon→GEO反馈\n    \n    # 默认拒绝\n    (\"*\", \"*\", \"*\"): False\n}\n```\n\n---\n\n## 六、Trace 协议\n\n### 6.1 TraceContext\n\n```python\n@dataclass\nclass TraceContext:\n    trace_id: str         # 64位UUID，贯穿整条请求链路\n    span_id: str          # 8位十六进制，当前操作ID\n    parent_span_id: Optional[str] = None  # 父span\n    \n    # 传播字段 (HTTP Header: traceparent)\n    # traceparent: 00-{trace_id}-{span_id}-{flags}\n    # flags: 01=采样, 00=不采样\n\n# ── Span生命周期 ──────────────────────────────────────\n# 开始: span_id = generate_span_id()\n# 进行中: 记录 start_time + 关键事件\n# 结束: 记录 end_time + status + attributes\n# 导出: 异步写入 SQLite + JSONL.gz\n```\n\n### 6.2 Trace传播示例\n\n```\n用户: \"帮我分析某产品在北美市场的机会\"\n  │\n  ▼\n[trace_id: abc123] chief-of-staff.intent-recognition (span: 00000001)\n  │\n  ├──▶ [并行执行组 1]\n  │     ├─ geo-01-market-research (span: 00000002, parent: 00000001)\n  │     │      ├─ sup-01-data-collect (span: 00000003, parent: 00000002) ✓\n  │     │      └─ [结果: 市场容量报告] (耗时: 2.3s)\n  │     │\n  │     └─ amz-01-product-select (span: 00000004, parent: 00000001)\n  │            ├─ sup-01-data-collect (span: 00000005, parent: 00000004) ✓\n  │            └─ [结果: 选品分析] (耗时: 2.8s)\n  │\n  └──▶ [等待依赖完成] → geo-03-content-strategy (span: 00000006, parent: 00000001)\n         ├─ geo-04-multilingual (span: 00000007, parent: 00000006) ✓\n         ├─ sup-02-content-gen (span: 00000008, parent: 00000006) ✓\n         └─ [结果: 内容策略报告] (耗时: 1.5s)\n  \n  ▼\n  chief-of-staff.result-aggregation (span: 00000009, parent: 00000001)\n  ├─ sup-05-report-gen (span: 00000010) ✓\n  └─ [最终报告] (总耗时: 8.2s)\n```\n\n---\n\n## 七、API接口\n\n### 7.1 Agent请求接口\n\n```yaml\nPOST /api/v1/agent/execute\nContent-Type: application/json\nAuthorization: Bearer {jwt}\n\nRequest:\n{\n  \"sender\": \"chief-of-staff\",\n  \"receiver\": \"geo-01-market-research\",\n  \"msg_type\": \"REQUEST\",\n  \"action\": \"market_research\",\n  \"priority\": 1,\n  \"payload\": {\n    \"product\": \"无线蓝牙耳机\",\n    \"regions\": [\"北美\", \"欧盟\", \"东南亚\"],\n    \"date_range\": \"2024-01-01~2026-03-31\"\n  },\n  \"trace_id\": \"abc123-def456\",\n  \"span_id\": \"00000001\",\n  \"ttl_seconds\": 300,\n  \"idempotency_key\": \"req-20260414-001\"\n}\n\nResponse (200 OK):\n{\n  \"msg_id\": \"msg-uuid-xxx\",\n  \"trace_id\": \"abc123-def456\",\n  \"status\": \"SUCCESS\",\n  \"result\": {\n    \"market_size\": \"TAM=$12.5B, SAM=$3.2B, SOM=$480M\",\n    \"top_regions\": [\"北美\", \"东南亚\", \"中东\"],\n    \"growth_rate\": \"+18% YoY\",\n    \"quality_score\": 87.5\n  },\n  \"quality_score\": 87.5,\n  \"metadata\": {\n    \"execution_time_ms\": 2314,\n    \"tokens_used\": 4521,\n    \"agent_id\": \"geo-01-market-research\"\n  }\n}\n```\n\n### 7.2 批量请求接口\n\n```yaml\nPOST /api/v1/agent/batch-execute\nAuthorization: Bearer {jwt}\n\nRequest:\n{\n  \"execution_mode\": \"PARALLEL\",  # PARALLEL / SEQUENTIAL / DAG\n  \"tasks\": [\n    {\"receiver\": \"geo-01\", \"action\": \"market_research\", \"payload\": {...}},\n    {\"receiver\": \"amz-01\", \"action\": \"product_select\", \"payload\": {...}},\n    {\"receiver\": \"sup-01\", \"action\": \"data_collect\", \"payload\": {...}}\n  ],\n  \"dependencies\": [\n    {\"from\": \"geo-01\", \"to\": \"geo-03\", \"condition\": \"always\"}\n  ],\n  \"max_parallel\": 10\n}\n\nResponse (200 OK):\n{\n  \"batch_id\": \"batch-xxx\",\n  \"execution_plan\": [...],  # DAG可视化\n  \"results\": {\n    \"geo-01\": {\"status\": \"SUCCESS\", \"result\": {...}},\n    \"amz-01\": {\"status\": \"SUCCESS\", \"result\": {...}},\n    \"sup-01\": {\"status\": \"PENDING\", \"estimated_start\": \"2026-04-14T10:00:00Z\"}\n  }\n}\n```\n\n---\n\n## 八、健康检查与监控\n\n### 8.1 Agent健康检查\n\n```python\n# /health/{agent_id} 接口响应格式\n{\n    \"agent_id\": \"geo-01-market-research\",\n    \"status\": \"HEALTHY\",  # HEALTHY / DEGRADED / DOWN\n    \"version\": \"1.0.0\",\n    \"uptime_seconds\": 86400,\n    \"metrics\": {\n        \"requests_total\": 1523,\n        \"requests_success\": 1498,\n        \"requests_failed\": 25,\n        \"avg_latency_ms\": 2340,\n        \"p95_latency_ms\": 5100,\n        \"error_rate\": 0.016,\n        \"queue_depth\": 3,\n        \"active_connections\": 12\n    },\n    \"dependencies\": {\n        \"sup-01-data-collect\": \"HEALTHY\",\n        \"sup-08-memory\": \"HEALTHY\"\n    },\n    \"last_heartbeat\": \"2026-04-14T10:00:05Z\"\n}\n```\n\n### 8.2 告警规则\n\n```yaml\nalerts:\n  - name: \"agent_down\"\n    condition: \"status == DOWN\"\n    severity: \"CRITICAL\"\n    channels: [\"slack\", \"email\", \"pagerduty\"]\n    \n  - name: \"high_error_rate\"\n    condition: \"error_rate > 0.05\"\n    severity: \"WARNING\"\n    channels: [\"slack\"]\n    \n  - name: \"high_latency\"\n    condition: \"p95_latency > 10000\"\n    severity: \"WARNING\"\n    channels: [\"slack\"]\n    \n  - name: \"queue_overflow\"\n    condition: \"queue_depth > 50\"\n    severity: \"CRITICAL\"\n    channels: [\"slack\", \"pagerduty\"]\n    \n  - name: \"security_event\"\n    condition: \"event_type == UNAUTHORIZED_ACCESS\"\n    severity: \"CRITICAL\"\n    channels: [\"security-team\", \"pagerduty\"]\n```\n\n---\n\n*本协议规范为 M-A3 30-Agent集群 的通信标准*\n*版本 v1.0 | 2026-04-14*\n\nFile v3.0.4:30-expansion/IMPLEMENTATION_PLAN.md\n\n# M-A3 30-Agent集群 分阶段实施计划\n> 版本：v1.0 | 日期：2026-04-14 | 作者：M-A3 幕僚长\n\n---\n\n## 一、实施策略\n\n### 1.1 核心原则\n\n| 原则 | 说明 |\n|------|------|\n| **先基础设施后业务** | Tier 0/1 Agent（调度/安全/记忆/数据）必须优先完成 |\n| **先核心后扩展** | P0 Agent优先，P1次之，P2最后 |\n| **并行开发** | 各域可同时开发，互不阻塞 |\n| **可测试交付** | 每个Agent交付前必须通过质量基准测试 |\n| **平滑灰度** | 新Agent逐步引入，不影响现有业务流程 |\n\n### 1.2 阶段总览\n\n```\nWeek 1-2   Phase 0: 基础设施层（10个Agent）\nWeek 3-4   Phase 1: GEO域 + 亚马逊域核心（10个Agent）\nWeek 5-6   Phase 2: GEO域 + 亚马逊域扩展（10个Agent）\nWeek 7-8   Phase 3: 全链路集成 + 压力测试\nWeek 9+    Phase 4: 持续优化 + 自进化增强\n```\n\n---\n\n## 二、详细实施计划\n\n### 📦 Phase 0：基础设施层（第1-2周）\n\n**目标**：建立集群底座，确保后续Agent可靠运行\n\n| # | Agent ID | Agent名称 | 优先级 | 预估工时 | 交付物 | 验收标准 |\n|---|---------|---------|--------|---------|--------|---------|\n| 1 | sup-09-scheduler | 调度协调Agent | P0 | 3天 | task_scheduler.py + DAG引擎 | 10个任务并行调度测试通过 |\n| 2 | sup-10-security | 安全审计Agent | P0 | 3天 | security_audit.py + RBAC矩阵 | PII检测测试10/10通过 |\n| 3 | sup-01-data-collect | 数据采集Agent | P0 | 3天 | data_collector.py + 适配器 | 3个数据源接入测试通过 |\n| 4 | sup-08-memory | 记忆管理Agent | P1 | 2天 | memory_manager.py + RAG | 记忆检索准确率>85% |\n| 5 | sup-02-content-gen | 内容生成Agent | P0 | 3天 | content_generator.py + 模板 | 批量生成质量评分>75分 |\n| 6 | sup-04-compliance | 合规检查Agent | P0 | 2天 | compliance_checker.py | 10条合规规则测试通过 |\n| 7 | sup-03-translation | 翻译Agent | P1 | 2天 | translator.py + 术语库 | 中英翻译质量评分>80 |\n| 8 | sup-07-quality | 质量评分Agent | P1 | 2天 | quality_scorer.py | 与人工评分相关性>0.85 |\n| 9 | geo-07-knowledge-graph | 知识图谱Agent | P0 | 3天 | knowledge_graph.py + JSON-LD | 实体识别准确率>90% |\n| 10 | amz-09-keywords | 关键词Agent | P1 | 2天 | keyword_agent.py + 词库 | 关键词覆盖率>竞品1.5倍 |\n\n**Phase 0 关键技术决策**：\n```\n✅ 调度引擎：基于Weighted Least Connections算法\n✅ 安全架构：RBAC × Agent × 操作类型\n✅ 数据管道：适配器模式，支持API/爬虫/文件\n✅ 记忆系统：SQLite FTS5 + RAG语义检索\n```\n\n**Phase 0 风险评估**：\n| 风险 | 概率 | 影响 | 缓解措施 |\n|------|------|------|---------|\n| 调度引擎并发死锁 | 中 | 高 | Phase 0第1个交付，深度测试 |\n| 安全Agent误报阻塞正常请求 | 低 | 高 | 灰度10%流量，逐步放量 |\n| 数据采集被反爬 | 中 | 中 | 多IP池 + 请求间隔 |\n\n---\n\n### 📊 Phase 1：核心业务Agent（第3-4周）\n\n**目标**：交付GEO域和亚马逊域的核心P0 Agent，形成业务闭环\n\n#### GEO域核心（5个）\n\n| # | Agent ID | Agent名称 | 优先级 | 预估工时 | 交付物 | 验收标准 |\n|---|---------|---------|--------|---------|--------|---------|\n| 1 | geo-01-market-research | 市场研究Agent | P0 | 3天 | market_researcher.py + 报告模板 | TAM/SAM/SOM三层模型输出正确 |\n| 2 | geo-03-content-strategy | 内容策略Agent | P0 | 3天 | content_strategist.py + 日历生成器 | 季度内容日历覆盖>200个话题 |\n| 3 | geo-08-intent-prediction | 意图预测Agent | P1 | 2天 | intent_predictor.py | 意图分类准确率>85%（对标94.3%目标） |\n| 4 | geo-02-competitor | 竞品分析Agent | P1 | 2天 | competitor_analyzer.py | GEO三维度评分可输出 |\n| 5 | geo-06-monitoring | 效果监测Agent | P1 | 2天 | geo_monitor.py + 告警引擎 | AI搜索引用率追踪可用 |\n\n#### 亚马逊域核心（5个）\n\n| # | Agent ID | Agent名称 | 优先级 | 预估工时 | 交付物 | 验收标准 |\n|---|---------|---------|--------|---------|--------|---------|\n| 1 | amz-01-product-select | 选品分析Agent | P0 | 3天 | product_selector.py + 利润模型 | 选品报告覆盖Top10候选 |\n| 2 | amz-02-listing | Listing优化Agent | P0 | 3天 | listing_optimizer.py | Listing评分>85分 |\n| 3 | amz-03-profit | 利润优化Agent | P0 | 3天 | profit_optimizer.py + 决策树 | 预测准确率>80% |\n| 4 | amz-04-ads | 广告投放Agent | P0 | 3天 | ads_manager.py + 竞价引擎 | ACOS优化建议可执行 |\n| 5 | amz-05-inventory | 库存管理Agent | P1 | 2天 | inventory_manager.py | 补货计划表生成正确 |\n\n**Phase 1 关键技术决策**：\n```\n✅ GEO意图预测：引入参考PureblueAI的意图分类模型\n✅ 利润优化：使用ProfitOptimizer决策树算法\n✅ 广告投放：基于TACOS导向的Bid调节算法\n```\n\n**Phase 1 业务闭环验证**：\n```\n用户输入：某消费电子新品 → \nGEO域：市场研究 → 竞品分析 → 内容策略 → 效果监测\n亚马逊域：选品分析 → Listing优化 → 利润优化 → 广告投放\n覆盖：产品上市前全链路\n```\n\n---\n\n### 🚀 Phase 2：扩展Agent（第5-6周）\n\n**目标**：交付所有剩余Agent，达到30个Agent完整覆盖\n\n#### GEO域扩展（5个）\n\n| # | Agent ID | Agent名称 | 优先级 | 预估工时 | 交付物 | 验收标准 |\n|---|---------|---------|--------|---------|--------|---------|\n| 1 | geo-04-multilingual | 多语言优化Agent | P1 | 2天 | multilingual_optimizer.py | 5个语种本地化完成 |\n| 2 | geo-05-platform-adapt | 平台适配Agent | P1 | 2天 | platform_adapter.py | 知乎/CSDN/LinkedIn适配 |\n| 3 | geo-09-schema | Schema优化Agent | P2 | 2天 | schema_optimizer.py | 全站Schema覆盖率>95% |\n| 4 | geo-10-regional | 地域策略Agent | P2 | 2天 | regional_strategist.py | 4个区域差异化策略 |\n| 5 | (整合Phase1) | - | - | - | GEO域工具链完善 | 全链路集成测试通过 |\n\n#### 亚马逊域扩展（5个）\n\n| # | Agent ID | Agent名称 | 优先级 | 预估工时 | 交付物 | 验收标准 |\n|---|---------|---------|--------|---------|--------|---------|\n| 1 | amz-06-review | 评价分析Agent | P1 | 2天 | review_analyzer.py + 情感分析 | 情感分析准确率>80% |\n| 2 | amz-07-competitor-monitor | 竞品监控Agent | P1 | 2天 | competitor_monitor.py | 5个ASIN监控可用 |\n| 3 | amz-08-pricing | 定价策略Agent | P1 | 2天 | pricing_strategist.py | 动态定价规则可配置 |\n| 4 | amz-10-reporting | 报表分析Agent | P2 | 2天 | reporting_agent.py | 日/周/月报自动生成 |\n| 5 | (整合Phase1) | - | - | - | 亚马逊域工具链完善 | 全链路集成测试通过 |\n\n#### 支撑域扩展（2个）\n\n| # | Agent ID | Agent名称 | 优先级 | 预估工时 | 交付物 | 验收标准 |\n|---|---------|---------|--------|---------|--------|---------|\n| 1 | sup-05-report-gen | 报告生成Agent | P1 | 2天 | report_generator.py + PPT模板 | Markdown/PDF/HTML/PPT四格式 |\n| 2 | sup-06-customer-service | 客户服务Agent | P2 | 2天 | customer_service.py + 对话引擎 | FAQ自动回复准确率>85% |\n\n**Phase 2 关键技术决策**：\n```\n✅ 多语言优化：文化适配引擎（非纯翻译）\n✅ 竞品监控：实时爬虫 + 价格告警阈值\n✅ 报告生成：ECharts可视化集成\n```\n\n---\n\n### 🔧 Phase 3：全链路集成 + 压力测试（第7-8周）\n\n**目标**：验证30个Agent的协同工作能力，确保生产级稳定性\n\n#### 3.1 集成测试\n\n| 测试场景 | 描述 | 预期结果 | 优先级 |\n|---------|------|---------|--------|\n| **GEO全链路** | 市场研究→竞品分析→内容策略→多语言→平台适配→效果监测 | 完整报告生成 | P0 |\n| **亚马逊全链路** | 选品→Listing→利润→广告→库存→定价→报表 | 完整运营方案 | P0 |\n| **跨域协作** | GEO市场研究 → 亚马逊选品（数据互通） | 选品报告包含GEO数据 | P0 |\n| **P2P通信** | 10组跨Agent消息传递 | 消息可靠到达 | P0 |\n| **降级测试** | Claude MA不可用 → Local → DeepSeek | 降级不影响核心功能 | P1 |\n| **并发压测** | 50个并发请求，30个Agent | 无死锁，响应时间P95<5s | P0 |\n\n#### 3.2 压力测试指标\n\n```python\nstress_test_scenarios = {\n    \"normal_load\": {\n        \"concurrent_users\": 50,\n        \"avg_tasks_per_user\": 3,\n        \"expected_p95_latency_ms\": 5000,\n        \"expected_error_rate\": 0.01\n    },\n    \"peak_load\": {\n        \"concurrent_users\": 200,\n        \"avg_tasks_per_user\": 5,\n        \"expected_p95_latency_ms\": 15000,\n        \"expected_error_rate\": 0.05\n    },\n    \"spike_load\": {\n        \"concurrent_users\": 500,\n        \"avg_tasks_per_user\": 3,\n        \"expected_p95_latency_ms\": 30000,\n        \"expected_error_rate\": 0.10,\n        \"expected_queue_time_s\": 120\n    }\n}\n```\n\n#### 3.3 安全压力测试\n\n```\nMCP协议漏洞扫描（43%风险覆盖）：\n- 命令注入测试：防止恶意prompt注入\n- 路径遍历测试：防止文件访问越界\n- 凭证泄露测试：防止API Key暴露\n- SSRF测试：防止内部服务探测\n覆盖率目标：100%工具函数通过扫描\n```\n\n---\n\n### 🌱 Phase 4：持续优化 + 自进化（第9周+）\n\n**目标**：让集群具备自我优化能力\n\n#### 4.1 自进化机制\n\n```\nlearnings/\n├── 2026-04-agent-failures/      # 失败经验\n├── 2026-04-agent-successes/      # 成功规则\n├── 2026-04-30-expansion/         # 扩展经验（本案）\n└── agent_capability_matrix.md   # Agent能力矩阵\n\n进化流程：\n1. SUP-07质量评分持续监测\n2. 低于基准触发learnings记录\n3. SUP-08记忆管理归类整理\n4. 定期（周）知识编译→Wiki\n5. 月度Agent能力画像更新\n```\n\n#### 4.2 性能优化计划\n\n| 优化项 | 目标 | 负责人 | 优先级 |\n|--------|------|-------|--------|\n| 调度算法优化 | P95延迟从5s→3s | SUP-09 | P1 |\n| 缓存命中率提升 | 从30%→60% | SUP-08 | P1 |\n| 意图预测模型升级 | 准确率85%→90% | GEO-08 | P1 |\n| 成本优化 | Token消耗降低20% | 全部 | P2 |\n\n#### 4.3 下一步扩展方向（2026 Q2-Q3）\n\n| 方向 | Agent数量 | 说明 |\n|------|----------|------|\n| 供应链域 | 5个 | 采购/物流/生产/质检/溯源 |\n| 客服域 | 5个 | 多语言客服/投诉处理/退款/工单/回访 |\n| 财务域 | 5个 | 成本分析/税务/汇率/预算/审计 |\n| **潜在新增** | **15个** | 45个Agent集群 |\n\n---\n\n## 三、交付物清单\n\n### 3.1 文档交付物\n\n| 文件路径 | 说明 | 阶段 | 状态 |\n|---------|------|------|------|\n| `30-expansion/30-agents-design.md` | 30个Agent完整设计方案 | Phase 0 | ✅ 已完成 |\n| `30-expansion/agent_registry.yaml` | Agent注册表配置 | Phase 0 | ✅ 已完成 |\n| `30-expansion/IMPLEMENTATION_PLAN.md` | 分阶段实施计划 | Phase 0 | ✅ 已完成 |\n| `30-expansion/agent_protocol.md` | Agent通信协议规范 | Phase 0 | 📋 待交付 |\n| `30-expansion/test_benchmark.md` | Agent质量基准测试集 | Phase 1 | 📋 待交付 |\n\n### 3.2 代码交付物\n\n```\nagent-cluster/\n├── agents/\n│   ├── geo/\n│   │   ├── geo-01-market-research/\n│   │   ├── geo-02-competitor/\n│   │   ├── geo-03-content-strategy/\n│   │   ├── geo-04-multilingual/\n│   │   ├── geo-05-platform-adapt/\n│   │   ├── geo-06-monitoring/\n│   │   ├── geo-07-knowledge-graph/\n│   │   ├── geo-08-intent-prediction/\n│   │   ├── geo-09-schema/\n│   │   └── geo-10-regional/\n│   ├── amazon/\n│   │   ├── amz-01-product-select/\n│   │   ├── amz-02-listing/\n│   │   ├── amz-03-profit/\n│   │   ├── amz-04-ads/\n│   │   ├── amz-05-inventory/\n│   │   ├── amz-06-review/\n│   │   ├── amz-07-competitor-monitor/\n│   │   ├── amz-08-pricing/\n│   │   ├── amz-09-keywords/\n│   │   └── amz-10-reporting/\n│   └── support/\n│       ├── sup-01-data-collect/\n│       ├── sup-02-content-gen/\n│       ├── sup-03-translation/\n│       ├── sup-04-compliance/\n│       ├── sup-05-report-gen/\n│       ├── sup-06-customer-service/\n│       ├── sup-07-quality/\n│       ├── sup-08-memory/\n│       ├── sup-09-scheduler/\n│       └── sup-10-security/\n```\n\n---\n\n## 四、里程碑\n\n| 里程碑 | 日期 | 交付内容 | 成功标准 |\n|--------|------|---------|---------|\n| M1 | Week 2末 | Phase 0完成 | 10个基础设施Agent可用 |\n| M2 | Week 4末 | Phase 1完成 | 20个Agent，覆盖核心业务 |\n| M3 | Week 6末 | Phase 2完成 | 30个Agent全部就绪 |\n| M4 | Week 8末 | Phase 3完成 | 全链路测试通过，压力测试达标 |\n| M5 | Week 10末 | Phase 4上线 | 自进化机制运行，知识库积累 |\n\n---\n\n## 五、资源预算\n\n### 5.1 开发工作量\n\n| 阶段 | Agent数量 | 开发天数 | 人力投入 |\n|------|----------|---------|---------|\n| Phase 0 | 10 | 14天 | 2人并行 |\n| Phase 1 | 10 | 14天 | 2人并行 |\n| Phase 2 | 10 | 14天 | 2人并行 |\n| Phase 3 | 集成 | 14天 | 2人并行 |\n| **总计** | **30** | **56天** | **约4人月** |\n\n### 5.2 成本估算\n\n| 成本项 | 月度成本 | 说明 |\n|--------|---------|------|\n| Claude MA API | ¥2,000-5,000 | 按量付费，Phase 0-1高消耗 |\n| DeepSeek API | ¥500-1,000 | Phase 0-2消耗 |\n| 云服务器 | ¥1,000-2,000 | 4核8G起步 |\n| 数据存储 | ¥200-500 | SQLite + OSS |\n| **月度合计** | **¥3,700-8,500** | |\n\n---\n\n*本计划由 M-A3 幕僚长 制定*\n*版本 v1.0 | 2026-04-14*\n*下一步行动：启动 Phase 0 开发*\n\nFile v3.0.4:api_integration/deepseek_integration.md\n\n# DeepSeek V3.2 API 集成文档\n\n> 为 DeepSeek V4 切换做准备 · 2026-04-14\n\n---\n\n## 1. 概述\n\n### 1.1 什么是 DeepSeek V3.2\n\nDeepSeek V3.2 是国产大模型 DeepSeek 的最新版本，通过 **OpenAI 兼容 API** 提供服务。与 V3.1 相比，V3.2 主要升级：\n\n| 特性 | V3.1 | V3.2 |\n|------|------|------|\n| 模型 ID | `deepseek-chat` | `deepseek-chat`（底层升级） |\n| 推理模型 ID | `deepseek-reasoner` | `deepseek-reasoner`（底层升级） |\n| 上下文窗口 | 128K | 128K |\n| `reasoning_content` 暴露 | ✅ | ✅（增强） |\n| JSON Mode | ✅ | ✅ |\n| Function Calling | ✅ | ✅ |\n| Beta 端点 8K max_tokens | ✅ | ✅ |\n| FIM Completion | ✅ | ✅（Beta） |\n| Chat Prefix Completion | ✅ | ✅（Beta） |\n| Context Caching | ❌ | ✅（降成本） |\n\n> **V4 预告**：DeepSeek V4 预计 2026 年 4 月下旬发布，届时只需修改 `model` 参数为 `deepseek-chat-v4` 或 `deepseek-reasoner-v4`，无需改动代码。\n\n### 1.2 两个模型的区别\n\n| 模型 ID | 模式 | 适用场景 | 特点 |\n|---------|------|----------|------|\n| `deepseek-chat` | 非思考模式 | 通用对话、代码生成、快速问答 | 响应快、成本低 |\n| `deepseek-reasoner` | 思考模式 | 数学推理、复杂分析、代码调试 | 内置思维链（CoT），返回 `reasoning_content` |\n\n两者底层共享 V3.2 架构，API 格式完全一致，**仅切换 `model` 参数即可切换模式**。\n\n---\n\n## 2. 快速开始\n\n### 2.1 安装依赖\n\n```bash\npip install httpx       # 推荐（已在 agent-cluster 间接依赖）\npip install openai      # 可选，OpenAI SDK 方式调用时使用\n```\n\n### 2.2 获取 API Key\n\n1. 访问 [https://platform.deepseek.com/api_keys](https://platform.deepseek.com/api_keys)\n2. 创建新 API Key（格式：`dsk-xxx`）\n3. 将 Key 存入环境变量（**不要硬编码**）\n\n```bash\n# Linux / macOS\nexport DEEPSEEK_API_KEY=\"dsk-your-key-here\"\n\n# Windows (PowerShell)\n$env:DEEPSEEK_API_KEY=\"dsk-your-key-here\"\n```\n\n### 2.3 Python 快速调用\n\n#### 方式一：直接使用 httpx（推荐用于 agent-cluster）\n\n```python\nimport httpx\nimport os\n\nclient = httpx.AsyncClient(timeout=120.0)\nresponse = await client.post(\n    \"https://api.deepseek.com/chat/completions\",\n    json={\n        \"model\": \"deepseek-chat\",\n        \"messages\": [\n            {\"role\": \"system\", \"content\": \"You are a helpful assistant.\"},\n            {\"role\": \"user\", \"content\": \"Explain quantum computing in one sentence.\"},\n        ],\n        \"max_tokens\": 500,\n        \"temperature\": 0.3,\n    },\n    headers={\n        \"Authorization\": f\"Bearer {os.environ['DEEPSEEK_API_KEY']}\",\n        \"Content-Type\": \"application/json\",\n    },\n)\ndata = response.json()\nprint(data[\"choices\"][0][\"message\"][\"content\"])\n```\n\n#### 方式二：OpenAI SDK 兼容调用\n\n```python\nfrom openai import OpenAI\n\nclient = OpenAI(\n    api_key=os.environ[\"DEEPSEEK_API_KEY\"],\n    base_url=\"https://api.deepseek.com\",  # 或 \"https://api.deepseek.com/v1\"\n)\n\nresponse = client.chat.completions.create(\n    model=\"deepseek-chat\",\n    messages=[\n        {\"role\": \"system\", \"content\": \"You are a helpful assistant.\"},\n        {\"role\": \"user\", \"content\": \"Hello!\"},\n    ],\n)\nprint(response.choices[0].message.content)\n```\n\n#### 方式三：cURL 快速测试\n\n```bash\ncurl https://api.deepseek.com/chat/completions \\\n  -H \"Content-Type: application/json\" \\\n  -H \"Authorization: Bearer $DEEPSEEK_API_KEY\" \\\n  -d '{\n    \"model\": \"deepseek-chat\",\n    \"messages\": [\n      {\"role\": \"system\", \"content\": \"You are a helpful assistant.\"},\n      {\"role\": \"user\", \"content\": \"Hello!\"}\n    ],\n    \"max_tokens\": 500,\n    \"temperature\": 0.3,\n    \"stream\": false\n  }'\n```\n\n---\n\n## 3. 引擎集成架构\n\n### 3.1 在 agent-cluster 中使用\n\n```python\nfrom execution.deepseek_engine import DeepSeekEngine\n\n# 初始化引擎\nengine = DeepSeekEngine({\n    \"api_key\": os.environ[\"DEEPSEEK_API_KEY\"],\n    \"model\": \"deepseek-chat\",       # 或 \"deepseek-reasoner\"\n    \"max_tokens\": 4096,\n    \"temperature\": 0.3,\n    \"json_mode\": False,\n})\n\n# 执行任务\nresult = await engine.execute(\n    task=\"查询今日库存情况\",\n    context={\n        \"user_id\": \"u001\",\n        \"user_role\": \"admin\",\n        \"intent_type\": \"stock_query\",\n    },\n)\n\nprint(result.output[\"content\"])\nprint(f\"Tokens: {result.tokens_used}, Latency: {result.latency_ms}ms\")\n```\n\n### 3.2 通过 EngineRouter 自动路由\n\nDeepSeek 引擎已注册到 `EngineRouter`，以下场景会自动路由到 DeepSeek：\n\n```python\nfrom execution.engine_router import EngineRouter, RoutingContext\n\nrouter = EngineRouter()\n\nctx = RoutingContext(\n    task=\"需要使用国产合规方案\",\n    intent_type=\"compliance\",\n    user_role=\"admin\",\n    scene=\"compliance\",\n    entities={},\n)\n# → 自动路由到 domestic-deepseek-chat\n```\n\n### 3.3 配置项说明\n\n| 参数 | 类型 | 默认值 | 说明 |\n|------|------|--------|------|\n| `api_key` | str | env: `DEEPSEEK_API_KEY` | API Key，优先级：config > env |\n| `model` | str | `deepseek-chat` | 模型 ID：`deepseek-chat` / `deepseek-reasoner` |\n| `base_url` | str | `https://api.deepseek.com` | API 端点 |\n| `max_tokens` | int | 4096 | 最大输出 token（Beta 端点上限 8K） |\n| `temperature` | float | 0.3 | 采样温度（0~1） |\n| `reasoning_effort` | float | 0.5 | 思维链努力程度（0~1，仅 R1 生效） |\n| `timeout` | float | 120.0 | 请求超时秒数 |\n| `json_mode` | bool | False | 强制 JSON 输出 |\n| `use_beta` | bool | False | 使用 Beta 端点（8K max_tokens / FIM / Prefix Completion） |\n\n---\n\n## 4. V3.2 高级特性\n\n### 4.1 reasoning_content（推理思维链）\n\n**仅 `deepseek-reasoner` 模式有效**\n\n```python\nresult = await engine.execute(\n    task=\"求 x³ - 6x² + 11x - 6 = 0 的解\",\n    context=build_context(intent_type=\"analysis\"),\n)\n\n# 最终答案\nprint(result.output[\"content\"])\n\n# 推理过程（V3.2 新增增强）\nif result.output.get(\"reasoning_content\"):\n    print(\"=== 推理过程 ===\")\n    print(result.output[\"reasoning_content\"])\n```\n\n**输出结构：**\n```json\n{\n  \"choices\": [{\n    \"message\": {\n      \"content\": \"答案是 42\",           // 最终答案\n      \"reasoning_content\": \"逐步推理...\" // 思维链（reasoner 模式）\n    }\n  }]\n}\n```\n\n### 4.2 JSON Mode（结构化输出）\n\n适用于需要程序化解析结果的场景：\n\n```python\nengine = DeepSeekEngine({\n    \"api_key\": api_key,\n    \"json_mode\": True,\n})\n\nresult = await engine.execute(\n    task='返回一个 JSON：{\"name\": str, \"age\": int, \"skills\": list[str]}',\n    context=build_context(),\n)\n\nimport json\ndata = json.loads(result.output[\"content\"])\nprint(data[\"name\"])\n```\n\n### 4.3 多轮对话（带历史）\n\n```python\nresult = await engine.execute(\n    task=\"继续\",\n    context={\n        **build_context(),\n        \"history\": [\n            {\"role\": \"user\", \"content\": \"你好\"},\n            {\"role\": \"assistant\", \"content\": \"你好！有什么可以帮你的？\"},\n        ],\n    },\n)\n```\n\n### 4.4 流式输出（SSE）\n\n```python\nasync for chunk in engine.stream(\"写一篇短文\", build_context()):\n    if chunk.done:\n        print(\"\\n[流式输出完成]\")\n    else:\n        print(chunk.content, end=\"\", flush=True)\n```\n\n> ⚠️ **注意**：`deepseek-reasoner` 模式暂不支持 SSE 流式输出，引擎会自动降级为阻塞调用后分词流式展示。\n\n### 4.5 Beta 端点（8K max_tokens / FIM / Prefix Completion）\n\n```python\nengine = DeepSeekEngine({\n    \"api_key\": api_key,\n    \"use_beta\": True,        # 切换到 /beta 端点\n    \"max_tokens\": 8192,      # 自动限制到 8K\n})\n```\n\nBeta 端点额外支持：\n- **FIM Completion**：代码中间补全（`POST /completions`）\n- **Chat Prefix Completion**：续写指定前缀\n\n### 4.6 Context Caching（降低长上下文成本）\n\n将长文档作为 context 复用时，DeepSeek V3.2 支持上下文缓存：\n\n```python\n# 目前通过 system prompt 注入长文档（未来版本将支持专用 cache API）\nmessages = [\n    {\"role\": \"system\", \"content\": \"参考文档：\\n\" + long_document},\n    {\"role\": \"user\", \"content\": \"基于上述文档回答：...\"},\n]\n```\n\n---\n\n## 5. 错误处理\n\n### 5.1 错误码对照表\n\n| HTTP 状态码 | error_code | 含义 | 处理建议 |\n|-------------|-----------|------|----------|\n| 401 | `invalid_api_key` | API Key 无效 | 检查 `DEEPSEEK_API_KEY` 配置 |\n| 403 | `forbidden` | 无访问权限 | 检查账户余额和权限 |\n| 429 | `rate_limit_exceeded` | 请求频率超限 | 添加重试延迟（指数退避） |\n| 500 | `internal_server_error` | DeepSeek 服务器错误 | 稍后重试 |\n| 503 | `service_unavailable` | 服务不可用 | 降级到其他引擎 |\n\n### 5.2 引擎内错误处理\n\n```python\nfrom execution.deepseek_engine import DeepSeekEngine, DeepSeekAPIError\n\nengine = DeepSeekEngine({\"api_key\": api_key})\n\ntry:\n    result = await engine.execute(task, context)\n    if not result.success:\n        logger.error(f\"DeepSeek 执行失败: {result.error}\")\n        # → 触发 EngineRouter 降级\nexcept DeepSeekAPIError as e:\n    if e.status_code == 429:\n        await asyncio.sleep(5)  # 退避重试\n        ...\n```\n\n### 5.3 降级策略\n\n在 `engines.yaml` 中配置：\n\n```yaml\nfallback:\n  order:\n    - local-self-built      # DeepSeek 失败 → 本地引擎\n    - claude-managed-agents # 本地也失败 → Claude MA\n  conditions:\n    timeout_seconds: 60\n    max_retries: 1\n    retry_on_error: true\n```\n\n---\n\n## 6. V4 切换指南\n\nV4 发布后，切换步骤：\n\n### 步骤 1：更新配置\n\n**`config/engines.yaml`**：\n```yaml\nengines:\n  deepseek:\n    model: \"deepseek-chat-v4\"      # V4 模型 ID（TBD，以官方公告为准）\n    # deepseek-reasoner-v4\n```\n\n或通过环境变量：\n```bash\nexport DEEPSEEK_MODEL=\"deepseek-chat-v4\"\n```\n\n### 步骤 2：运行回归测试\n\n```bash\nDEEPSEEK_API_KEY=sk-xxx pytest agent-cluster/tests/test_deepseek_engine.py -v\n```\n\n### 步骤 3：验证生产场景\n\n重点验证：\n- [ ] 基础对话（deepseek-chat-v4）\n- [ ] 复杂推理（deepseek-reasoner-v4）\n- [ ] JSON Mode 输出格式\n- [ ] Function Calling（Agent 场景）\n- [ ] 响应延迟和 Token 消耗对比\n\n### 步骤 4：监控指标\n\n```python\n# 切换后监控指标\n- result.latency_ms（延迟）\n- result.tokens_used（Token 消耗）\n- result.success（成功率）\n- error rate（按错误码分类）\n```\n\n---\n\n## 7. 安全注意事项\n\n1. **API Key 安全存储**\n   - ✅ 使用环境变量或 `.env` 文件\n   - ❌ 禁止硬编码在代码中\n   - ❌ 禁止提交到 Git（已在 `.gitignore` 中排除 `.env`）\n\n2. **生产环境白名单**\n   ```python\n   # 生产环境建议启用 httpx 白名单\n   export DEEPSEEK_WHITELIST_ENABLED=\"true\"\n   # 域名 api.deepseek.com 已在白名单中\n   ```\n\n3. **日志脱敏**\n   - API Key 在日志中显示为 `dsk-xxx***`\n   - 请求内容不记录 Token 字段\n\n---\n\n## 8. 性能基准参考\n\n> 以下为参考值，实际性能取决于网络和任务复杂度。\n\n| 模型 | 场景 | 预期延迟 | Token 消耗 |\n|------|------|----------|------------|\n| `deepseek-chat` | 简单对话（100字） | 0.5~2s | ~200 tokens |\n| `deepseek-chat` | 代码生成（200字） | 1~3s | ~400 tokens |\n| `deepseek-reasoner` | 数学推理（中等） | 3~8s | ~800 tokens（含 CoT） |\n| `deepseek-reasoner` | 复杂分析（长） | 8~15s | ~1500 tokens（含 CoT） |\n\n---\n\n## 9. 参考资源\n\n| 资源 | 地址 |\n|------|------|\n| DeepSeek 官方文档 | https://api-docs.deepseek.com/ |\n| DeepSeek 控制台（API Key 管理） | https://platform.deepseek.com/api_keys |\n| V3.2 模型说明 | https://platform.deepseek.com/docs |\n| DeepSeek V4 发布公告 | 待更新（预计 2026-04 下旬） |\n\n---\n\n## Change Log\n\n| 日期 | 版本 | 变更内容 |\n|------|------|----------|\n| 2026-04-14 | v0.1 | 初始文档，V3.2 API 集成测试版本 |\n| 2026-04-（待更新） | v0.2 | V4 切换记录 |\n\nFile v3.0.4:cms_executor/ARCHITECTURE.md\n\n# CMS Executor — Architecture Design\n\n> **Gradial GEO 时代的 Agent 集群能力扩展**  \n> 将\"洞察→执行\"闭环从分析层推进到 CMS 直连写入层，让现有 30 个 Agent 拥有直接操作电商/建站平台的能力。\n\n---\n\n## 1. 系统全景\n\n```\n┌─────────────────────────────────────────────────────────────────┐\n│                    M-A3 Agent Cluster (30 agents)                │\n│  ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐           │\n│  │ GEO      │ │ Amazon   │ │ Content  │ │ Finance  │   ...     │\n│  │ Analyst  │ │ Operator │ │ Creator  │ │ Agent    │           │\n│  └────┬─────┘ └────┬─────┘ └────┬─────┘ └────┬─────┘           │\n│       └─────────── │ ──────────│────────────┘                 │\n│                     ▼                                           │\n│          ┌─────────────────────┐                               │\n│          │   CMS Executor API  │  ← agent_integration.py       │\n│          │  (MCP Protocol)    │                               │\n│          └─────────┬───────────┘                               │\n│                    │                                            │\n│   ┌────────────────┼──────────────────────────────────────┐   │\n│   │                ▼                                          │   │\n│   │  ┌─────────────────────────────────────────────────┐    │   │\n│   │  │              Execution Engine                   │    │   │\n│   │  │  ┌─────────┐  ┌──────────┐  ┌──────────────┐  │    │   │\n│   │  │  │Executor │  │Approval  │  │ Rollback     │  │    │   │\n│   │  │  │         │←→│ Workflow │←→│ Manager      │  │    │   │\n│   │  │  └────┬────┘  └──────────┘  └──────────────┘  │    │   │\n│   │  │       │                                    │    │   │\n│   │  │       ▼                                    │    │   │\n│   │  │  ┌─────────┐  ┌──────────┐  ┌────────────┐  │    │   │\n│   │  │  │ Audit  │  │Safety   │  │ Version   │  │    │   │\n│   │  │  │ Logger │  │Sandbox  │  │ Snapshot  │  │    │   │\n│   │  │  └─────────┘  └──────────┘  └────────────┘  │    │   │\n│   │  └─────────────────────────────────────────────────┘    │   │\n│   │                     │                                  │   │\n│   └─────────────────────┼──────────────────────────────────┘   │\n│                         ▼                                        │\n│   ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐          │\n│   │WordPress │ │ Shopify  │ │ Amazon   │ │ Magento  │  Custom  │\n│   │Connector │ │Connector │ │Connector │ │Connector │  Sites   │\n│   │ (REST)   │ │(GraphQL) │ │ (SP-API) │ │ (REST)   │          │\n│   └──────────┘ └──────────┘ └──────────┘ └──────────┘          │\n└─────────────────────────────────────────────────────────────────┘\n```\n\n---\n\n## 2. 核心设计原则\n\n| 原则 | 说明 |\n|------|------|\n| **零信任写入** | 所有 CMS 写入必须经过审批流程（手动/自动） |\n| **幂等性优先** | 所有写操作支持幂等执行，防止重复提交 |\n| **快照回滚** | 每次变更前自动创建可恢复快照 |\n| **Agent 沙箱** | Agent 调用 CMS 能力前必须在安全沙箱中预演 |\n| **全链路审计** | 每次操作产生不可篡改的审计记录 |\n| **渐进式暴露** | 能力按 Agent 角色分级暴露（read → write → admin） |\n\n---\n\n## 3. CMS 连接器抽象层\n\n### 3.1 连接器基类 (`connectors/base_connector.py`)\n\n```\nBaseCMSConnector (ABC)\n├── platform: str              # \"wordpress\" | \"shopify\" | \"amazon\" | \"magento\" | \"custom\"\n├── credentials: Credentials     # 加密存储的凭据\n├── capabilities: list[str]    # [\"create_post\", \"update_post\", \"delete_post\", ...]\n├── api_base: str             # API endpoint base URL\n│\n├── async connect()            # 建立连接（带连接池复用）\n├── async disconnect()         # 断开连接\n├── async health_check()       # 健康检查\n│\n├── async read(resource_id)   # 读取单个资源\n├── async list(filters)        # 列表查询\n├── async create(data)         # 创建资源（幂等 Key 防止重复）\n├── async update(resource_id, data)   # 更新资源（先快照）\n├── async delete(resource_id)  # 删除资源（软删除优先）\n│\n├── async snapshot(resource_id)  # 创建变更前快照\n├── async rollback(snapshot_id)   # 从快照恢复\n│\n├── to_cms_operation(op)      # 将统一操作转换为平台原生格式\n└── normalize_response(resp)   # 归一化响应格式\n```\n\n### 3.2 平台能力矩阵\n\n| 操作 | WordPress | Shopify | Amazon SP-API | Magento | Custom |\n|------|-----------|---------|---------------|---------|--------|\n| 读取内容 | ✅ | ✅ | ✅ | ✅ | ✅ |\n| 创建内容 | ✅ | ✅ | ✅ | ✅ | ✅ |\n| 更新内容 | ✅ | ✅ | ✅ | ✅ | ✅ |\n| 删除内容 | ✅ | ✅ | ✅ | ✅ | ✅ |\n| 上传媒体 | ✅ | ✅ | ✅ | ✅ | ✅ |\n| 更新库存 | - | ✅ | ✅ | ✅ | ✅ |\n| 更新价格 | - | ✅ | ✅ | ✅ | ✅ |\n| SEO 元数据 | ✅ | ✅ | ✅ | ✅ | ✅ |\n\n---\n\n## 4. 执行引擎 (`engine/`)\n\n### 4.1 执行编排器 (`executor.py`)\n\n```\nCMSTaskExecutor\n│\n├── execute(plan, context)    # 主入口，编排完整执行流程\n│   ├── 1. validate_plan()    # 验证执行计划合法性\n│   ├── 2. check_sandbox()    # Agent 沙箱预演（高风险操作）\n│   ├── 3. submit_approval()  # 提交审批（根据风险等级）\n│   ├── 4. await_approval()   # 等待审批（同步/异步）\n│   ├── 5. create_snapshot()   # 创建变更前快照\n│   ├── 6. execute_ops()      # 执行写操作（幂等保护）\n│   ├── 7. verify_result()    # 验证执行结果\n│   └── 8. emit_audit()       # 记录审计日志\n│\n├── execute_batch(plans)      # 批量执行（并行/串行）\n├── preview(plan)             # 仅预览，不实际执行\n└── cancel(execution_id)      # 取消正在执行的任务\n```\n\n### 4.2 审批工作流 (`approval.py`)\n\n```\nApprovalWorkflow\n│\n├── RiskLevel 枚举: LOW / MEDIUM / HIGH / CRITICAL\n│\n├── determine_risk_level(op)  # 基于操作类型+数据量+目标平台评估风险\n│\n├── ApprovalChain\n│   ├── LOW:     Auto-Approve（记录即放行）\n│   ├── MEDIUM:  Content-Creator Agent 自审\n│   ├── HIGH:    Chief-of-Staff 审批\n│   └── CRITICAL: 人工介入（发送通知）\n│\n├── submit(plan, risk_level)  # 提交审批\n├── approve(execution_id)     # 批准\n├── reject(execution_id, reason)  # 拒绝\n├── escalate(execution_id)    # 升级\n└── get_status(execution_id) # 查询审批状态\n│\n├── 审批超时: MEDIUM=10min, HIGH=1h, CRITICAL=24h\n└── 超时处理: 自动降级或通知\n```\n\n### 4.3 回滚机制 (`rollback.py`)\n\n```\nRollbackManager\n│\n├── SnapshotStore            # 快照存储（文件系统/S3）\n│   ├── snapshots/           # 按平台/日期组织\n│   │   ├── wordpress/\n│   │   ├── shopify/\n│   │   └── amazon/\n│\n├── create_snapshot(op)      # 创建变更前快照\n│   ├── 资源当前状态 JSON\n│   ├── 操作元数据（时间/执行人/Agent）\n│   └── 变更指纹（SHA256）\n│\n├── rollback(snapshot_id)    # 从快照恢复\n│   ├── 验证快照完整性\n│   ├── 确认回滚目标仍存在\n│   ├── 执行逆向操作\n│   └── 验证恢复结果\n│\n├── list_snapshots(filters)  # 列出可用快照\n├── compare_versions(id1, id2) # 对比两个版本差异\n└── auto_cleanup(retention_days=30)  # 自动清理过期快照\n```\n\n### 4.4 审计日志 (`audit.py`)\n\n```\nCMSAuditLogger\n│\n├── 继承自 agent-cluster/safety/audit_logger.py 的 AuditLogger\n│\n├── 扩展事件类型:\n│   ├── CMS_CONNECT        # CMS 连接建立\n│   ├── CMS_DISCONNECT     # CMS 连接断开\n│   ├── CMS_READ           # 内容读取\n│   ├── CMS_WRITE          # 内容写入（核心事件）\n│   ├── CMS_APPROVAL       # 审批动作\n│   ├── CMS_ROLLBACK       # 回滚操作\n│   └── CMS_SANDBOX        # 沙箱执行\n│\n├── log_cms_operation(op)  # 结构化记录 CMS 操作\n│   ├── platform, resource_id, operation_type\n│   ├── before_snapshot_id, after_snapshot_id\n│   ├── risk_level, approval_status\n│   └── agent_id, execution_id\n│\n├── generate_compliance_report()  # 生成合规报告\n└── export_audit_trail(start, end)  # 导出审计轨迹\n```\n\n---\n\n## 5. Agent 集成 (`agent_integration.py`)\n\n### 5.1 调用接口\n\n```python\n# 方式 1: 直接调用（通过 Python import）\nfrom agent_integration import CMSExecutorClient\n\nclient = CMSExecutorClient(agent_id=\"geo_analyst_01\")\nresult = await client.execute(\n    platform=\"wordpress\",\n    operation=\"update_post\",\n    resource_id=\"post_12345\",\n    data={\"content\": \"Updated SEO content...\"},\n    agent_context={\"agent_id\": \"geo_analyst_01\", \"intent\": \"seo_optimization\"}\n)\n\n# 方式 2: MCP Protocol（通过 MCP Gateway）\n# MCP 工具: cms_execute, cms_preview, cms_rollback, cms_health\n```\n\n### 5.2 角色权限矩阵\n\n| Agent 角色 | READ | WRITE | ADMIN | 特殊权限 |\n|------------|------|-------|-------|---------|\n| geo_analyst | ✅ | ✅ (preview) | ❌ | cms_preview |\n| content_creator | ✅ | ✅ (own content) | ❌ | cms_preview |\n| amazon_operator | ✅ | ✅ (inventory/price) | ❌ | amazon specific |\n| chief_of_staff | ✅ | ✅ | ✅ | cms_rollback |\n| admin | ✅ | ✅ | ✅ | ALL |\n\n### 5.3 与 GEO MCP 连接器的集成\n\n```\nGEO Intent → GEO Analyst Agent\n              ↓\n    Gradial GEO Engine（分析/生成）\n              ↓\n    CMS Executor Client（执行）\n              ↓\n    CMS Executor API → Approval → CMS Connector\n```\n\n---\n\n## 6. 安全设计\n\n### 6.1 四层安全模型\n\n```\nLayer 1 — Authentication（认证）\n  └── API Key / OAuth Token（平台级别加密存储）\n\nLayer 2 — Authorization（授权）\n  └── RBAC: Agent 角色 → CMS 能力映射表\n\nLayer 3 — Safety Sandbox（安全沙箱）\n  └── 高风险操作（delete/mass_update）在沙箱中预演\n  └── 危险关键词检测（\"DROP TABLE\"、\"rm -rf\"）\n\nLayer 4 — Audit Logging（全链路审计）\n  └── 所有操作不可篡改记录\n  └── SOC 2 合规报告\n```\n\n### 6.2 危险操作黑名单\n\n```\n🚫 DELETE 整站 / 全量删除\n🚫 UPDATE price/quantity 为 0（无下限保护）\n🚫 修改管理员账户凭据\n🚫 执行任意 SQL / 代码注入\n🚫 批量覆盖（非增量更新）\n```\n\n---\n\n## 7. 文件结构\n\n```\nagent-cluster/cms-executor/\n├── ARCHITECTURE.md                    # 本文档\n├── __init__.py\n├── agent_integration.py               # Agent 调用接口 + MCP 暴露\n│\n├── connectors/\n│   ├── __init__.py\n│   ├── base_connector.py              # ABC 连接器基类\n│   ├── wordpress_connector.py         # WordPress REST API\n│   ├── shopify_connector.py           # Shopify GraphQL\n│   ├── amazon_connector.py            # Amazon SP-API\n│   └── magento_connector.py           # Magento REST\n│\n├── engine/\n│   ├── __init__.py\n│   ├── executor.py                    # 任务编排引擎\n│   ├── approval.py                    # 审批流程\n│   ├── rollback.py                    # 快照回滚\n│   └── audit.py                       # CMS 专用审计\n│\n└── tests/\n    ├── __init__.py\n    ├── test_base_connector.py\n    ├── test_wordpress_connector.py\n    ├── test_shopify_connector.py\n    ├── test_executor.py\n    ├── test_approval.py\n    └── test_rollback.py\n```\n\n---\n\n## 8. 依赖关系\n\n```\nagent-cluster/\n├── safety/audit_logger.py   ← CMSAuditLogger 继承\n├── error_handling/          ← 异常处理复用\n├── execution/circuit_breaker.py  ← 熔断保护\n└── quality/gate.py          ← 质量门禁复用\n```\n\n---\n\n## 9. 验收标准\n\n- [ ] 4 个平台连接器（WordPress/Shopify/Amazon/Magento）全部实现\n- [ ] 执行引擎支持同步/异步审批\n- [ ] 回滚机制在 5 秒内完成单资源恢复\n- [ ] 审计日志覆盖 100% 的写操作\n- [ ] 沙箱预演对 Agent 透明，执行时间 < 2 秒\n- [ ] 所有测试通过 pytest（≥ 85% 覆盖率）\n- [ ] MCP 协议正确暴露 CMS 工具集\n\nFile v3.0.4:memory/ARCHITECTURE.md\n\n# M-A3 多Agent记忆层架构设计\n\n> 跨 Agent 知识共享与协同记忆系统 | agent-cluster/memory/\n\n---\n\n## 一、设计目标\n\n| 目标 | 实现方式 |\n|------|---------|\n| 会话记忆同步 | SessionSync + JSONL 事件日志 |\n| Agent 私有记忆隔离 | PrivateMemory + SQLite scope 隔离 |\n| 共享知识池 | SharedKnowledgePool + 权限矩阵 |\n| 持久化存储检索 | PersistentStore + FTS5 全文索引 |\n\n---\n\n## 二、架构概览\n\n```\n┌─────────────────────────────────────────────────────────────┐\n│                    外部调用方（Agent / Orchestrator）           │\n└──────────────────────────┬──────────────────────────────────┘\n                           │\n                           ▼\n┌──────────────────────────────────────────────────────────────┐\n│              MemoryRouter（统一入口/路由层）                     │\n│  · 自动 scope 判断   · 权限校验   · 路由分发                   │\n└───────┬─────────────────┬──────────────────┬────────────────┘\n        │                 │                  │\n        ▼                 ▼                  ▼\n┌───────────────┐ ┌─────────────────┐ ┌────────────────────┐\n│ PrivateMemory │ │SharedKnowledgePool│ │   SessionSync      │\n│ (Agent私有隔离)│ │  (跨Agent共享)   │ │  (会话同步+事件流)   │\n└───────┬───────┘ └────────┬────────┘ └────────┬─────────────┘\n        │                  │                   │\n        ▼                  ▼                   ▼\n┌──────────────────────────────────────────────────────────────┐\n│                 PersistentStore（SQLite 持久化）                │\n│  ┌──────────────────┐  ┌──────────────────────────────────┐ │\n│  │  WAL 模式高并发   │  │  FTS5 全文搜索索引                 │ │\n│  │  乐观锁版本控制   │  │  软删除 + 自动清理触发器            │ │\n│  └──────────────────┘  └──────────────────────────────────┘ │\n│  ┌──────────────┐  ┌──────────────┐  ┌──────────────────┐   │\n│  │ memory.db     │  │sessions.db   │  │session_events.jsonl│  │\n│  │ (记忆主库)    │  │ (会话元数据) │  │ (事件流日志)       │  │\n│  └──────────────┘  └──────────────┘  └──────────────────┘   │\n└──────────────────────────────────────────────────────────────┘\n                           │\n                           ▼\n┌──────────────────────────────────────────────────────────────┐\n│                  MemoryIndex（检索增强层）                      │\n│  · 多信号加权重排   · 相似记忆发现   · 热门标签   · 知识图谱   │\n└──────────────────────────────────────────────────────────────┘\n```\n\n---\n\n## 三、三层记忆作用域\n\n### 3.1 Private（私有记忆）\n- 每个 Agent 独立的存储空间\n- 权限：`agent_id == self`（仅自己可读/写）\n- 内容：个人偏好、专用知识、提炼规则、任务中间状态\n- TTL：由重要性自动决定（CRITICAL=1年，EPHEMERAL=1小时）\n\n### 3.2 Shared（共享知识池）\n- 所有 Agent 可读，授权 Agent 可写\n- 权限矩阵：\n  ```\n  orchestrator  → 可写全部\n  admin         → 可写全部\n  其他Agent     → 仅可读\n  ```\n- 池子类型：\n  - `shared_facts` - 共享事实（库存状态、汇率等）\n  - `collaboration_artifacts` - 协作成果（分析报告、决策文档）\n  - `org_knowledge` - 组织知识（政策、业务规则）\n  - `global_rules` - 全局协作协议\n\n### 3.3 Session（会话记忆）\n- 仅当前会话参与者可见\n- 会话结束后 24 小时自动过期\n- 内容：事件流、中间结果共享、决策广播\n\n---\n\n## 四、文件结构\n\n```\nagent-cluster/memory/\n├── __init__.py              # 包导出\n├── memory_core.py           # 核心数据模型（MemoryEntry, Query, Result）\n├── persistent_store.py      # SQLite 持久化（WAL + FTS5 + 触发器）\n├── private_memory.py        # Agent 私有记忆（隔离读写）\n├── shared_knowledge.py      # 共享知识池（权限矩阵）\n├── session_sync.py          # 会话同步（事件流 + 快照）\n├── memory_index.py          # 全文检索（加权重排 + 知识图谱）\n├── memory_router.py          # 统一路由入口\n├── memory_api.py             # FastAPI REST 接口（可选）\n├── memory_integration.py    # 集成胶水（Agent记忆胶水/会话恢复/协同）\n└── demo.py                   # 使用演示\n```\n\n---\n\n## 五、核心接口\n\n### 5.1 统一写入 `MemoryRouter.memorize()`\n```python\nentry_id = mr.memorize(\n    content=\"SKU001 库存预警：仅剩 50 件\",\n    agent_id=\"inventory\",\n    scope=MemoryScope.SHARED,          # 或 \"private\" / \"session\" / \"auto\"\n    memory_type=MemoryType.FACT,\n    importance=MemoryImportance.HIGH,\n    tags=[\"inventory\", \"alert\"],\n    related_agent_ids=[\"finance\", \"procurement\"],\n)\n```\n\n### 5.2 统一查询 `MemoryRouter.recall()`\n```python\nresult = mr.recall(\n    agent_id=\"finance\",\n    query_text=\"库存\",\n    scope=\"shared\",\n    limit=20,\n)\nfor entry in result.entries:\n    print(entry.content)\n```\n\n### 5.3 Agent 记忆胶水 `AgentMemoryGlue`\n```python\nglue = AgentMemoryGlue(\"inventory\")\n\n# 装饰器：自动记忆任务结果\n@glue.memorize_outcome\ndef query_inventory(sku):\n    return db.query(sku)\n\n# 快捷方法\nglue.remember_preference(\"优先仓库\", \"华东\")\nglue.remember_knowledge(\"爆款规则：日销 > 30 件需补货\")\nglue.remember_rule(\"安全水位 = 3天 × 日均销量\")\n\n# 跨Agent同步\nglue.share_with(\n    content=\"库存预警：SKU001 仅剩 50 件\",\n    target_agent_ids=[\"procurement\", \"finance\"],\n)\n```\n\n### 5.4 会话恢复 `SessionRecovery`\n```python\nrecovery = SessionRecovery(\"inventory\")\n\n# 保存检查点\nrecovery.save_checkpoint(session_id, task_state={\"step\": 2, \"data\": result})\n\n# 加载检查点\nstate = recovery.load_checkpoint(session_id)\n```\n\n### 5.5 协同记忆 `CollaborationMemory`\n```python\ncollab = CollaborationMemory(\"orchestrator\")\n\n# 启动协作\nsession_id = collab.start_collaboration(\n    task_id=\"月度报告\",\n    participants=[\"finance\", \"procurement\", \"sales\"],\n)\n\n# 记录决策（自动提议为全局规则）\ncollab.record_collaboration_decision(\n    session_id=session_id,\n    decision=\"毛利率目标 35%\",\n    decided_by=\"finance\",\n)\n```\n\n---\n\n## 六、权限矩阵\n\n| 操作 | Agent自身 | 其他Agent | Orchestrator |\n|------|----------|----------|-------------|\n| 读私有记忆 | ✅ | ❌ | ✅ |\n| 写私有记忆 | ✅ | ❌ | ❌ |\n| 读共享知识 | ✅ | ✅ | ✅ |\n| 写共享知识 | ❌ | ❌ | ✅ |\n| 读会话记忆 | ✅（参与者）| ✅（参与者）| ✅ |\n| 写会话记忆 | ✅（参与者）| ✅（参与者）| ✅ |\n| 删除他人共享 | ❌ | ❌ | ✅ |\n\n---\n\n## 七、生命周期管理\n\n```\n记忆创建 → 访问计数 → TTL 倒计时\n                            ↓\n    ┌── 重要性 ≥ HIGH ──→ 保留长期\n    │\n    └── 重要性 < HIGH ──→ TTL 到期 ──→ 软删除 ──→ FTS触发器自动清理\n                                                    │\n                                            ┌── 超过30天 ──→ 最终删除\n                                            └── 30天内 ──→ 可恢复\n```\n\n---\n\n## 八、FastAPI 启动方式\n\n```python\nfrom agent_cluster.memory.api import create_api\nfrom agent_cluster.memory import MemoryRouter\n\nrouter = MemoryRouter()\napp = create_api(router)\n\n# uvicorn.run(app, host=\"0.0.0.0\", port=8081)\n```\n\nAPI 端点：\n- `POST /memory/memorize` - 存储记忆\n- `POST /memory/recall` - 查询记忆\n- `POST /knowledge/publish` - 发布共享知识\n- `POST /session/create` - 创建协作会话\n- `POST /sync` - 跨Agent知识同步\n- `GET /context/{agent_id}` - 构建Agent上下文\n\n---\n\n## 九、集成现有系统\n\n### 9.1 Orchestrator 集成\n```python\nfrom agent_cluster.memory.memory_integration import OrchestratorMemoryMixin\n\nclass Orchestrator(OrchestratorMemoryMixin, BaseOrchestrator):\n    def on_task_complete(self, task, result, agent_id):\n        self.memorize_task_outcome(\n            task=task.description,\n            agent_id=agent_id,\n            outcome=str(result),\n            success=result.status == \"success\",\n        )\n```\n\n### 9.2 专业 Agent 集成\n```python\nfrom agent_cluster.memory.memory_integration import AgentMemoryGlue\n\nclass InventoryAgent:\n    def __init__(self):\n        self.memory = AgentMemoryGlue(\"inventory\")\n    \n    def query(self, sku):\n        result = self._query_db(sku)\n        self.memory.remember_task(\"query_inventory\", str(result), True)\n        return result\n```\n\n---\n\n## 十、性能特性\n\n| 特性 | 实现 |\n|------|------|\n| 高并发写入 | SQLite WAL 模式 |\n| 全文检索 | FTS5（支持前缀匹配） |\n| 检索加速 | 重要性加权重排 |\n| 连接复用 | 单例连接池 |\n| 内存效率 | SessionContext 按需加载 |\n| 数据安全 | 软删除 + 乐观锁 |\n\nFile v3.0.4:memory/P0-OPTIMIZATION.md\n\n# Dreaming 蒸馏引擎 P0 优化 — 使用说明\n\n## 交付物\n\n| 文件 | 说明 |\n|------|------|\n| `agent-cluster/memory/immediate_skill_hook.py` | 即时 Skill 生成钩子 |\n| `agent-cluster/memory/fts_search.py` | FTS5 全文检索增强层 |\n| `agent-cluster/memory/tests/test_immediate_skill_hook.py` | 单元测试（22 tests） |\n| `agent-cluster/memory/tests/test_fts_search.py` | 单元测试（27 tests） |\n\n## P0-1：即时 Skill 生成钩子\n\n### 核心机制\n\n任务完成后立即触发，置信度分流：\n\n```\ntask_result\n  → extract_skill_candidate()    LLM提取（失败→规则兜底）\n  → confidence >= 0.8  → write_skill_document()   直接写盘\n  → confidence 0.5-0.8 → add_to_dreaming_queue()  入梦境队列\n  → confidence < 0.5  → discard\n```\n\n### 快速使用\n\n```python\nfrom memory.immediate_skill_hook import create_hook\n\n# 创建钩子\nhook = create_hook(agent_id=\"agent-001\")\n\n# 在任务执行完成后调用\ntask_result = await agent.execute(task)\nawait hook.after_task_complete(task_result)\n```\n\n### 高级配置\n\n```python\nfrom memory.immediate_skill_hook import ImmediateSkillHook\n\nhook = ImmediateSkillHook(\n    agent_id=\"agent-001\",\n    skills_dir=\"data/skills\",           # Skill 文档输出目录\n    dream_queue_path=\"data/dream_queue.jsonl\",  # 梦境队列路径\n    llm_callable=my_llm_func,          # LLM 提取函数（可选）\n    persist_to_memory=True,            # 同步写入记忆系统（供 FTS 检索）\n)\n\n# 自定义置信度阈值\nhook.HIGH_CONFIDENCE_THRESHOLD = 0.85\nhook.MEDIUM_CONFIDENCE_THRESHOLD = 0.6\n```\n\n### LLM 提取器（可选）\n\n```python\ndef my_llm(query: str) -> str:\n    \"\"\"你的 LLM 接口，返回 JSON 字符串\"\"\"\n    return llm_client.chat([{\"role\": \"user\", \"content\": query}])\n\nhook = ImmediateSkillHook(llm_callable=my_llm)\n```\n\n### Skill 文档格式（agentskills.io 兼容）\n\n```markdown\n# Fix_SSL_Error\n\n> Fix Python SSL certificate verification error\n\n## Metadata\n- **Version**: `1.0.0`\n- **Confidence**: `0.92` (LLM structured extraction)\n- **Extracted at**: 2026-04-16T00:00:00+00:00\n- **Source task**: `task-001`\n\n## Triggers\n- `how to fix`\n\n## Actions\n1. Update certificates\n2. Set REQUESTS_CA_BUNDLE environment variable\n```\n\n---\n\n## P0-2：FTS5 全文检索层\n\n### 核心能力\n\n- **纯 FTS5 BM25 排序**：零外部依赖，直接利用 SQLite 内置 BM25\n- **snippet 高亮**：`【关键词】` 风格（MaxHermes 同款）\n- **过滤**：scope、memory_type、importance、agent_id\n- **分页**：offset/limit\n- **去重**：按 entry_id 自动去重\n- **混合搜索**：可注入语义函数，融合 BM25 + 语义分\n\n### 快速使用\n\n```python\nfrom memory.fts_search import create_searcher\n\n# 纯 FTS5 搜索（推荐，无外部依赖）\nsearcher = create_searcher()\nresult = searcher.search(\"python ssl error\", limit=10)\n\nfor hit in result.hits:\n    print(hit.snippet)   # 【SSL】certificate...\n    print(hit.bm25_score)\n```\n\n### 过滤与分页\n\n```python\nresult = searcher.search(\n    \"deployment\",\n    scopes=[\"shared\"],              # 作用域过滤\n    memory_types=[\"procedure\"],     # 记忆类型过滤\n    min_importance=4,               # 最低重要性\n    agent_id=\"agent-001\",           # 指定 Agent\n    limit=5,\n    offset=10,                      # 第二页\n)\n```\n\n### 混合搜索\n\n```python\nfrom memory.fts_search import HybridSearcher\n\ndef my_embedding(query: str, texts: list[str]) -> list[float]:\n    \"\"\"你的 embedding 服务，返回 0-1 相似度分数列表\"\"\"\n    return embedding_client.similarity(query, texts)\n\nsearcher = HybridSearcher()\nresult = searcher.search(\n    \"async task handling\",\n    mode=\"keyword_first\",           # keyword_first | semantic_first | pure_fts\n    semantic_weight=0.4,            # 语义权重（0=纯关键词）\n    semantic_func=my_embedding,\n    limit=10,\n)\n```\n\n### HybridSearcher 三种模式\n\n| 模式 | 说明 |\n|------|------|\n| `pure_fts` | 纯 FTS5 BM25，无外部依赖（默认） |\n| `keyword_first` | FTS5 为主排序，语义作 boost |\n| `semantic_first` | 语义相似度为主，FTS5 候选兜底 |\n\n### snippet 高亮说明\n\nSQLite snippet() 默认返回 `...<b>term</b>...`，本模块使用 `【term】` 风格：\n\n```python\nresult = searcher.search(\"ssl\", highlight_open=\"【\", highlight_close=\"】\")\n# → \"【SSL】certificate verify failed\"\n```\n\n### 获取 Skill 候选（Dreaming 引擎用）\n\n```python\nfrom memory.fts_search import FTS5Searcher\n\nsearcher = FTS5Searcher()\nskill_candidates = searcher.get_skill_candidates(limit=20)\n# → [SearchHit(entry_id=..., is_skill_candidate=True, ...)]\n```\n\n---\n\n## 测试运行\n\n```bash\n# 运行全部测试\ncd agent-cluster\npytest memory/tests/ -v\n\n# 分别运行\npytest memory/tests/test_immediate_skill_hook.py -v\npytest memory/tests/test_fts_search.py -v\n```\n\n---\n\n## 与现有系统集成\n\n### 集成 immediate_skill_hook 到 Agent\n\n```python\nfrom memory.immediate_skill_hook import create_hook\n\nhook = create_hook(agent_id=\"agent-001\")\n\nasync def run_task(task):\n    result = await agent.execute(task)\n    # 任务完成后立即触发 Skill 提炼\n    await hook.after_task_complete(result)\n    return result\n```\n\n### 集成 fts_search 到 Dreaming 引擎\n\n```python\nfrom memory.fts_search import create_searcher\n\nsearcher = create_searcher()\n\n# Dreaming 引擎召回相关记忆（无需 recall signal）\nrelated = searcher.search(\n    query=\"previous error patterns\",\n    memory_types=[\"procedure\", \"episode\"],\n    min_importance=3,\n    limit=5,\n)\n```\n\n### 在 memory_api.py 中启用 FTS5 增强\n\n```python\nfrom memory.fts_search import create_searcher\n\n# 在 FastAPI 路由中使用\n@router.get(\"/search\")\nasync def search_memories(q: str, limit: int = 10):\n    searcher = create_searcher()\n    result = searcher.search(q, limit=limit)\n    return result.to_dict()\n```\n\nFile v3.0.4:PRICING.md\n\n# 外贸硅基军团 - 定价策略\n\n## 定价方案\n\n### Free（免费版）\n- **适合**：个人外贸SOHO、创业初期团队\n- **功能**：\n  - Shopify基础连接（单店铺）\n  - 库存查询\n  - 采购订单基础审批\n  - 4个Agent协作（库存/采购/财务/物流）\n  - 每日100次API调用\n  - 基础链路追踪\n\n### Pro - ¥9.9/月\n- **适合**：成长型外贸企业\n- **功能**：\n  - Shopify + WooCommerce + Magento全平台\n  - 多店铺管理（最多5个）\n  - 完整A2A多Agent协作\n  - 财务审核流\n  - 每日1000次API调用\n  - 高级链路追踪 + Mermaid可视化\n  - 邮件/飞书审批通知\n\n### Enterprise（企业版）\n- **适合**：大型跨境电商企业\n- **功能**：\n  - 无限店铺连接\n  - 自定义Agent工作流编排\n  - 企业级审批流配置\n  - 无限API调用\n  - 专属合规审核Agent\n  - 技术支持 + SLA保障\n  - 本地化部署选项\n\n## 价格优势\n- 比Salesforce Agentforce低80%（后者起步$25/用户/月）\n- 比定制开发快10倍上线\n- 无平台锁定，按月订阅随时取消\n\n---\n*最后更新：2026-04-17*\n\nArchive v3.0.3: 144 files, 521505 bytes\n\nFiles: __init__.py (406b), _meta.json (132b), 30-expansion/30-agents-design.md (51910b), 30-expansion/agent_protocol.md (16120b), 30-expansion/agent_registry.yaml (18892b), 30-expansion/IMPLEMENTATION_PLAN.md (13710b), api_integration/__init__.py (936b), api_integration/api_adapter.py (21482b), api_integration/api_config.py (9243b), api_integration/api_health.py (8629b), api_integration/deepseek_integration.md (11899b), api_integration/mock_data.py (14178b), cms_approvals/04cf553c-e0c9-4732-870b-83f96e443b52.json (533b), cms_approvals/1c9e92a7-46f3-4a6d-a0d9-59c60ff0fee1.json (692b), cms_approvals/265648cd-fd95-4b2c-a1e7-4051cbe771cc.json (569b), cms_approvals/2c363c59-d353-4e3d-a2b6-aa4e321fa278.json (569b), cms_approvals/31ff6cde-b148-47d1-980f-6953a7e1fe98.json (569b), cms_approvals/326cf306-f324-478f-9abb-952657fd8a58.json (569b), cms_approvals/380f9c96-b301-4cbc-9a61-d508323241bf.json (717b), cms_approvals/5709cd14-a912-4bb9-a877-e8fd3c029ec2.json (569b), cms_approvals/a2941003-ecb3-41bc-8522-088b9bae1f41.json (701b), cms_approvals/b260ecbf-649d-4da4-b064-3093114d7ca7.json (704b), cms_approvals/d55ab274-327d-4090-985d-8c8c4d795c5c.json (683b), cms_approvals/de294f4f-a55b-43b2-bbae-3840b81f2e85.json (674b), cms_approvals/eb03a9c1-ecb6-4007-ada6-aaefd49ef4ac.json (533b), cms_executor/__init__.py (2412b), cms_executor/agent_integration.py (26443b), cms_executor/ARCHITECTURE.md (14538b), cms_executor/connectors/__init__.py (891b), cms_executor/connectors/amazon_connector.py (14762b), cms_executor/connectors/base_connector.py (27345b), cms_executor/connectors/magento_connector.py (16667b), cms_executor/connectors/shopify_connector.py (16183b), cms_executor/connectors/wordpress_connector.py (11068b), cms_executor/engine/__init__.py (642b), cms_executor/engine/approval.py (11726b), cms_executor/engine/audit.py (17904b), cms_executor/engine/executor.py (19384b), cms_executor/engine/rollback.py (15292b), cms_executor/tests/__init__.py (224b), cms_executor/tests/test_agent_integration.py (14358b), cms_executor/tests/test_approval.py (8408b), cms_executor/tests/test_base_connector.py (12002b), cms_executor/tests/test_connectors.py (14832b), cms_executor/tests/test_executor.py (13274b), cms_executor/tests/test_rollback.py (9323b), cms-executor/connectors/__init__.py (457b), cms-executor/connectors/amazon_connector.py (26421b), cms-executor/connectors/base_connector.py (5652b), cms-executor/connectors/shopify_connector.py (25896b), cms-executor/connectors/wordpress_connector.py (8616b), cms-executor/engine/__init__.py (168b), cms-executor/engine/approval.py (9193b), cms-executor/engine/rollback.py (10602b), cms-executor/README.md (12573b), cms-executor/tests/test_cms_executor.py (13197b), cms-executor/tests/test_connectors.py (31593b), collaboration/__init__.py (1067b), collaboration/state_sync.py (10976b), collaboration/task_protocol.py (15311b), collaboration/tests/__init__.py (22570b), collaboration/trace_tracker.py (11782b), collaboration/workflow_engine.py (12650b), config/agents.yaml (3021b), config/engines.yaml (7322b), config/permissions.yaml (3718b), config/workflows.yaml (3799b), error_handling/__init__.py (1260b), error_handling/exception_middleware.py (11348b), error_handling/operation_log.py (15910b), error_handling/retry_policy.py (10861b), error_handling/task_state_machine.py (13609b), execution/__init__.py (1564b), execution/circuit_breaker.py (40704b), execution/claude_ma_engine.py (8921b), execution/deepseek_engine.py (37869b), execution/engine_base.py (23172b), execution/engine_router.py (37840b), execution/gpt6_engine.py (27838b), execution/local_engine.py (11985b)\n\nFile v3.0.3:SKILL.md\n\n---\nname: foreign-trade-silicon-army\ndescription: 亚马逊外贸B2B多CMS Agent协作系统。支持Shopify/WooCommerce/Magento三大平台，A2A架构协调库存/采购/财务/物流四大专家Agent，三层安全网保障审批合规。即装即用，零配置开箱。\nversion: 3.0.3\ntriggers:\n  - \"外贸\"\n  - \"CMS\"\n  - \"Shopify\"\n  - \"WooCommerce\"\n  - \"Magento\"\n  - \"库存\"\n  - \"采购\"\n  - \"Amazon\"\n  - \"B2B\"\ntags:\n  - agent\n  - a2a\n  - cms\n  - ecommerce\n  - b2b\n  - foreign-trade\n  - multi-agent\nrequires:\n  python_packages:\n    - httpx\n    - pyyaml\n    - fastapi\n    - uvicorn\n  env:\n    - SYSTEM_MODE\n    - SAP_BASE_URL\n    - SAP_API_KEY\n    - YONYOU_BASE_URL\n    - YONYOU_APPKEY\n    - KINGDEE_BASE_URL\n    - ANTHROPIC_API_KEY\n    - DEEPSEEK_API_KEY\nhomepage: https://github.com/WangM-A3/agent-cluster\nauthors:\n  - WangM-A3\npricing: free,pro=9.9,enterprise\n---\n\n# 产业互联网硅基军团 v2.0\n\n> 企业级Multi-Agent智能体集群系统，基于**1+N架构**（1个幕僚长+20个专业Agent），参考OpenClaw Main Agent、腾讯ADP Router设计。v2.0全面升级：真实ERP API接入、协作流程细化、错误处理体系。\n\n## 核心能力\n\n### v2.0三大升级\n\n**1. 真实API接入层（api_integration/）**\n- 多ERP适配器：SAP S/4HANA、用友U8/NC/YonBIP、金蝶K3 Cloud/EAS、通用REST\n- 断路器模式（Circuit Breaker）+ 故障自动降级\n- 健康检查轮询（10s间隔，自动摘除异常节点）\n- 演示模式保留（MockDataGenerator，variance=0.1随机波动）\n\n**2. 跨Agent协作细化（collaboration/）**\n- 细粒度任务协议（TaskMessage）：依赖声明、优先级、TTL\n- 状态同步（SharedStateManager）：TTL+pub/sub通知机制\n- 全链路追踪（CollaborationTracker）：trace_id/span_id + Mermaid时序图可视化\n\n**3. 错误处理与状态管理（error_handling/）**\n- 7状态任务状态机：pending→running→success/failed/retry/timeout/cancelled\n- 10类异常自动分类：VALIDATION/NETWORK/TIMEOUT/AUTH/RESOURCE/NOT_FOUND等\n- 5种重试策略：FIXED/EXPONENTIAL/FIBONACCI/JITTER/ADAPTIVE\n- 敏感信息脱敏 + SOC2合规审计日志\n\n## 系统架构\n\n```\n用户请求 → Orchestrator（意图识别→任务拆解→智能体调度）\n    ↓\n20个专业Agent：采购/生产/销售/财务/运营/战略/研发/人力/合规\n    ↓\nAPI Integration Layer（v2.0新增）\n  ├─ SAP/用友/金蝶适配器（真实ERP）\n  └─ 断路器+健康检查+Mock降级\n```\n\n## 目录结构\n\n```\nagent-cluster/\n├── orchestrator.py              # 指挥智能体（核心调度器）\n├── api_integration/              # v2.0新增：真实API接入层\n│   ├── api_adapter.py           # 多ERP适配器（SAP/用友/金蝶）\n│   ├── api_config.py            # 配置化管理\n│   ├── api_health.py            # 健康检查+断路器\n│   └── mock_data.py             # 模拟数据（开发/演示）\n├── collaboration/                # v2.0新增：跨Agent协作\n│   ├── task_protocol.py         # 细粒度任务协议\n│   ├── state_sync.py            # 状态同步+TTL+pub/sub\n│   ├── trace_tracker.py         # 全链路追踪+Mermaid\n│   └── workflow_engine.py       # 混合执行引擎\n├── error_handling/              # v2.0新增：错误处理\n│   ├── task_state_machine.py    # 7状态任务状态机\n│   ├── exception_middleware.py  # 统一异常处理\n│   ├── retry_policy.py          # 5种重试策略\n│   └── operation_log.py         # 操作日志+脱敏+合规\n├── specialists/                 # 专业智能体\n│   ├── inventory_agent.py      # 库存智能体\n│   ├── logistics_agent.py       # 物流智能体\n│   ├── procurement_agent.py     # 采购智能体\n│   ├── finance_agent.py         # 财务智能体\n│   └── doc_agent.py            # 工艺文档智能体\n├── mcp_servers/                 # MCP协议封装\n│   ├── erp_server.py           # ERP系统接口\n│   ├── wms_server.py           # WMS仓库管理接口\n│   └── srm_server.py           # SRM供应商管理接口\n├── safety/                      # 安全围栏\n│   ├── permission_manager.py    # RBAC权限管理\n│   ├── audit_logger.py          # 全链路审计日志\n│   └── human_loop.py            # 人机回环审批\n└── config/                      # 配置文件\n    ├── agents.yaml             # 智能体定义\n    ├── workflows.yaml          # 工作流配置\n    └── permissions.yaml        # 权限矩阵\n```\n\n## 快速开始\n\n### 环境要求\n- Python 3.10+\n- 依赖：`pip install pyyaml fastapi uvicorn httpx aiofiles`\n\n### 运行\n```bash\ncd agent-cluster\npython orchestrator.py\n```\n\n### 配置（生产模式）\n```bash\nexport SYSTEM_MODE=production\nexport SAP_BASE_URL=https://sap.example.com\nexport SAP_API_KEY=sk-xxx\nexport YONYOU_BASE_URL=https://yonyou.example.com\n```\n\n## 触发词\n\n塑化报价 | 塑料原料采购 | 库存管理 | 生产排产 | 客户跟进 | 供应商比价 | B2B运营 | 工厂管理 | ERP集成 | 智能客服 | 行业KPI | 成本核算 | 硅基军团 | 工业Agent | 制造业AI | 产业互联网\n\n## 标签\n\n制造业AI, 产业互联网, Multi-Agent, ERP集成, 智能排产, 供应商管理, 报价系统, 智能工厂, AI运营, 企业数字化\n\n## 分类\n\n效率工具\n\n## 版本\n\nv2.0.0 - 三大升级：真实API接入层、协作流程细化、错误处理与状态管理\n\nFile v3.0.3:cms-executor/README.md\n\n# CMS Executor - 多平台 CMS 连接器\n\n## 概述\n\n本模块提供多平台（WordPress / Shopify / Amazon）CMS 直连执行能力，支持通过平台原生 API 创建/更新/删除内容，配合审批引擎和回滚机制实现安全的内容管理。\n\n## 目录结构\n\n```\ncms-executor/\n├── connectors/\n│   ├── __init__.py              # 统一导出\n│   ├── base_connector.py        # 抽象基类（所有 CMS 平台通用接口）\n│   ├── wordpress_connector.py   # WordPress REST API 实现\n│   ├── shopify_connector.py     # Shopify GraphQL Admin API 实现\n│   └── amazon_connector.py      # Amazon SP-API 实现\n├── engine/\n│   ├── __init__.py\n│   ├── approval.py              # 多级审批引擎\n│   └── rollback.py              # 变更追踪与回滚\n└── tests/\n    ├── test_cms_executor.py     # WordPress & 审批引擎测试\n    └── test_connectors.py       # Shopify & Amazon 连接器测试\n```\n\n---\n\n## 平台支持\n\n| 平台 | API | 认证方式 |\n|------|-----|---------|\n| WordPress | REST API v2 | Application Passwords |\n| Shopify | GraphQL Admin API 2024-01 | OAuth / API Key |\n| Amazon | SP-API | LWA OAuth |\n\n---\n\n## 1. Shopify 连接器\n\n### 认证配置\n\n1. 在 Shopify 后台创建私有应用或自定义应用\n2. 配置 API 权限（Products, Variants, Images）\n3. 获取访问令牌（Access Token）\n\n### 基本使用\n\n```python\nfrom connectors import ShopifyConnector, CMSCredential, ContentPayload\n\n# 初始化连接器\ncred = CMSCredential(\n    url=\"https://your-shop.myshopify.com\",\n    api_key=\"shpat_xxxxxxxxxxxxx\",   # 访问令牌\n)\nshopify = ShopifyConnector(cred)\n\n# 验证连接\nif shopify.authenticate():\n    print(\"✅ Shopify 连接成功\")\n\n# ── 商品 CRUD ────────────────────────────────────────\nfrom connectors.shopify_connector import ProductPayload, VariantPayload\n\n# 方式一：使用 ContentPayload（兼容基类）\npayload = ContentPayload(title=\"My Product\", content=\"<p>Description</p>\", status=\"draft\")\nresult = shopify.create_content(payload)\nprint(f\"✅ 商品已创建，ID={result['id']}, handle={result['handle']}\")\n\n# 方式二：使用 ProductPayload（Shopify 原生字段）\nfrom connectors.shopify_connector import ProductPayload, VariantPayload, ImagePayload, SEOPayload\nproduct = ProductPayload(\n    title=\"Premium Widget\",\n    body_html=\"<p>High quality widget</p>\",\n    vendor=\"Acme Corp\",\n    product_type=\"Electronics\",\n    status=\"active\",\n    tags=[\"widget\", \"premium\"],\n    variants=[\n        VariantPayload(\n            title=\"Blue / Large\",\n            price=\"29.99\",\n            sku=\"WGT-BL-L\",\n            inventory_quantity=100,\n            option1=\"Blue\", option2=\"Large\",\n        ),\n        VariantPayload(\n            title=\"Red / Large\",\n            price=\"29.99\",\n            sku=\"WGT-RE-L\",\n            inventory_quantity=50,\n            option1=\"Red\", option2=\"Large\",\n        ),\n    ],\n    images=[\n        ImagePayload(src=\"https://cdn.example.com/widget.jpg\", alt_text=\"Widget front view\"),\n    ],\n    seo=SEOPayload(title=\"Premium Widget - Acme Corp\", description=\"Shop premium widgets\"),\n)\n# 通过内部 GraphQL API 直接创建\nresp = shopify._graphql(shopify._build_product_create_mutation(product))\n\n# ── 变体操作 ──────────────────────────────────────────\nvariant = VariantPayload(\n    title=\"Green / Medium\",\n    price=\"24.99\",\n    sku=\"WGT-GN-M\",\n    inventory_quantity=30,\n)\nshopify.create_variant(product_id=123, variant=variant)\nshopify.update_variant(variant_id=555, variant=VariantPayload(price=\"22.99\", inventory_quantity=10))\nshopify.delete_variant(variant_id=555)\n\n# ── 图片操作 ─────────────────────────────────────────\nfrom connectors.shopify_connector import ImagePayload\nshopify.add_product_image(123, ImagePayload(src=\"https://cdn.example.com/extra.jpg\"))\nshopify.delete_product_image(image_id=\"111\")\n\n# ── 商品列表 ─────────────────────────────────────────\nproducts = shopify.list_content({\"first\": 50, \"query\": \"status:active\"})\nfor p in products:\n    print(f\"  {p['id']} | {p['title']} | {p['handle']}\")\n\n# ── 删除商品（归档） ──────────────────────────────────\nshopify.delete_content(product_id=123)       # 归档（soft delete）\nshopify.delete_content(product_id=123, force=True)  # 永久删除\n\n# ── 历史 ─────────────────────────────────────────────\nhistory = shopify.get_history()\nfor h in history:\n    print(f\"  [{h['timestamp']}] {h['operation']} - {h['status']}\")\n\n# ── 回滚 ─────────────────────────────────────────────\n# 回滚最近一次创建操作\nfor rec in reversed(history):\n    if rec[\"operation\"] == \"create\" and rec[\"status\"] == \"executed\":\n        shopify.rollback_operation(rec[\"id\"])\n        print(\"✅ 回滚成功\")\n        break\n\nshopify.close()\n```\n\n---\n\n## 2. Amazon SP-API 连接器\n\n### 认证配置\n\n1. 在 Seller Central 注册 SP-API 应用，获取 Client ID 和 Client Secret\n2. 完成 LWA OAuth 流程，获取 Refresh Token\n3. Refresh Token 可长期有效，用于自动刷新 Access Token\n\n### 基本使用\n\n```python\nfrom connectors import AmazonConnector, CMSCredential, ContentPayload\n\n# 初始化连接器\ncred = CMSCredential(\n    url=\"https://sellercentral.amazon.com\",\n    username=\"amzn1.application-xxx.client_id\",\n    app_password=\"amzn1.application-xxx.client_secret\",\n    api_key=\"Atza|xxx-access-token\",  # 可选，已有 access token\n)\namazon = AmazonConnector(\n    cred,\n    region=\"na\",                    # na | eu | fe\n    marketplace_id=\"A1AM79NJPZON8\",  # 美国市场\n)\n\n# 设置 Refresh Token（用于自动续期 access token）\namazon.set_refresh_token(\"Atzr|...\")\n\n# 验证连接\nif amazon.authenticate():\n    print(\"✅ Amazon SP-API 连接成功\")\n\n# ── 商品列表操作 ─────────────────────────────────────\nfrom connectors.amazon_connector import AmazonListingPayload\n\n# 方式一：使用 ContentPayload\npayload = ContentPayload(title=\"USB-C Cable\", content=\"High-speed USB-C cable 2m\")\nresult = amazon.create_content(payload)\n\n# 方式二：使用 AmazonListingPayload（完整字段）\nlisting = AmazonListingPayload(\n    sku=\"USB-C-2M-BLK\",\n    product_type=\"ELECTRONIC_ACCESSORY\",\n    title=\"USB-C to USB-C Cable 2m Fast Charging\",\n    description=\"Braided nylon, 100W PD fast charge\",\n    brand=\"CableMax\",\n    manufacturer=\"CableMax Inc\",\n    price_amount=12.99,\n    price_currency=\"USD\",\n    quantity=500,\n    condition_type=\"New\",\n    fulfillment_channel=\"MFN\",\n    bullet_points=[\n        \"100W Power Delivery\",\n        \"USB 3.1 Gen 2, 10Gbps\",\n        \"Durable braided nylon\",\n    ],\n)\nresp = amazon._spapi_request(\n    \"PUT\",\n    f\"/listings/2021-08-01/items/{SELLER_ID}/{listing.sku}?marketplaceId={marketplace}\",\n    token=access_token,\n    data={\"attributes\": {...}},\n)\n\n# 更新商品\nresult = amazon.update_content(sku_int, payload)\n\n# 删除商品\nresult = amazon.delete_content(sku_int)\n\n# ── 库存管理 ──────────────────────────────────────────\n# 更新库存\namazon.update_inventory(sku=\"USB-C-2M-BLK\", quantity=300, fulfillment_channel=\"MFN\")\n# 查询库存\ninv = amazon.get_inventory(\"USB-C-2M-BLK\")\n# 批量查询\ninventories = amazon.list_inventory({\"next_token\": \"\"})\n\n# ── 价格管理 ──────────────────────────────────────────\n# 更新价格\namazon.update_pricing(sku=\"USB-C-2M-BLK\", amount=14.99, currency=\"USD\")\n# 查询价格\nprice = amazon.get_pricing(\"USB-C-2M-BLK\")\n\n# ── 报告 ─────────────────────────────────────────────\nfrom connectors.amazon_connector import PricePayload, InventoryPayload\n\n# 请求商品报告\nreport = amazon.request_report(\n    report_type=\"GET_MERCHANT_LISTINGS_DATA\",\n    marketplace_ids=[\"A1AM79NJPZON8\"],\n)\nprint(f\"Report ID: {report['report_id']}, Status: {report['status']}\")\n\n# 查询报告状态\nstatus = amazon.get_report(report[\"report_id\"])\n# 获取报告下载链接\ndocument = amazon.get_report_document(status[\"payload\"][\"reportDocumentId\"])\ndownload_url = document[\"url\"]  # 预签名 S3 URL\n\n# ── Feed 提交（批量 XML） ─────────────────────────────\nxml_content = '<?xml version=\"1.0\"?><AmazonEnvelope><Header>...</Header></AmazonEnvelope>'\nfeed = amazon.submit_feed(\n    feed_type=\"POST_PRODUCT_DATA\",\n    content=xml_content,\n    content_type=\"text/xml\",\n)\nprint(f\"Feed ID: {feed['feed_id']}, Status: {feed['status']}\")\n\n# ── 历史与回滚 ────────────────────────────────────────\nhistory = amazon.get_history()\nfor h in history:\n    print(f\"  [{h['timestamp']}] {h['operation']} {h['entity_type']} - {h['status']}\")\n\namazon.close()\n```\n\n---\n\n## 3. 审批流程（通用）\n\n```python\nfrom engine import ApprovalEngine, ApprovalRule, ApprovalLevel, OperationType\n\nengine = ApprovalEngine()\n\n# 创建审批请求（草稿 → 自动通过）\nreq = engine.create_request(\n    submitter=\"auto-bot\",\n    operation=OperationType.CREATE,\n    payload={\"title\": \"新商品\", \"content\": \"...\", \"status\": \"draft\"},\n)\n\nif engine.can_auto_approve(req):\n    engine.auto_approve(req)\n    print(\"✅ 自动审批通过\")\nelse:\n    print(f\"⏳ 等待审批，级别: {req.level.value}\")\n    engine.vote(req.id, approver=\"admin\", decision=\"approve\")\n```\n\n---\n\n## 4. 回滚操作\n\n```python\nfrom engine import RollbackEngine\n\nrollback = RollbackEngine(shopify)  # 或 AmazonConnector 实例\n\n# 查看最近变更\nchanges = rollback.list_recent_changes(hours=24)\nfor c in changes:\n    print(f\"  {c['id']} | {c['operation']} | {c['status']}\")\n\n# 生成回滚计划（不执行）\nplan = rollback.plan_rollback(record_id=\"abc12345\")\nprint(f\"回滚计划: {plan.strategy.value}, 步骤数: {len(plan.steps)}\")\n\n# 执行回滚\nresult = rollback.execute_rollback(plan.plan_id, force=True)\nprint(f\"回滚结果: {'成功' if result['success'] else '失败'}\")\n```\n\n---\n\n## 风险分级\n\n| 操作 | 触发条件 | 审批级别 |\n|------|---------|---------|\n| 创建商品 | 草稿状态 + 内容安全 | **AUTO_PASS** |\n| 更新商品 | 任意状态 | **SINGLE** |\n| 删除商品 | 任意状态 | **MULTI_SIGN** |\n| 价格变更 | 价格修改 | **MULTI_SIGN** |\n| 包含危险关键词 | `<script>`/`<?php` 等 | **MULTI_SIGN** |\n\n**AUTO_PASS**：无需审批，系统自动通过\n**SINGLE**：单人审批（任意一位审批人同意即可）\n**MULTI_SIGN**：多人会签（所有审批人全部同意）\n**ANY_SIGN**：任意一人同意即可\n\n---\n\n## 测试运行\n\n```bash\ncd agent-cluster/cms-executor\n\n# 运行 Shopify & Amazon 连接器测试（推荐）\npython -m pytest tests/test_connectors.py -v\n\n# 运行 WordPress & 审批引擎测试\npython -m pytest tests/test_cms_executor.py -v\n\n# 运行全部测试\npython -m pytest tests/ -v\n\n# 或直接运行\npython tests/test_connectors.py\npython tests/test_cms_executor.py\n```\n\n---\n\n## 扩展其他 CMS\n\n继承 `BaseCMSConnector` 抽象类，实现以下方法即可：\n\n```python\nfrom connectors.base_connector import BaseCMSConnector, CMSCredential, ContentPayload\n\nclass MyCMSConnector(BaseCMSConnector):\n    def authenticate(self) -> bool: ...\n    def create_content(self, payload: ContentPayload) -> Dict[str, Any]: ...\n    def update_content(self, content_id: int, payload: ContentPayload) -> Dict[str, Any]: ...\n    def delete_content(self, content_id: int, force: bool = False) -> Dict[str, Any]: ...\n    def get_content(self, content_id: int) -> Dict[str, Any]: ...\n    def list_content(self, params: Dict[str, Any] = None) -> List[Dict[str, Any]]: ...\n    def _do_rollback(self, record: OperationRecord) -> bool: ...\n```\n\nFile v3.0.3:README.md\n\n# 企业级智能体集群系统 v3.0\n\n> 基于 **1+N** 架构的智能体协作系统，参考 OpenClaw Main Agent、腾讯ADP Router、智己汽车研发设计集群\n\n## 核心升级（v3.0）\n\n本次更新将执行层解耦，支持**三引擎热切换**：\n\n1. **执行层抽象（ExecutionEngine）**：统一接口抽象，支持 Claude MA / Local / DeepSeek 三种引擎\n2. **EngineRouter 智能路由**：基于意图/场景/关键词的自动引擎选择，支持降级\n3. **向后兼容**：现有 `handle_request` API 完全不受影响，新增 `execute_with_engine` API\n\n## 核心升级（v2.0）\n\n本次更新根据用户评测反馈进行了三大改进：\n\n1. **真实API接入层**：适配器模式支持多ERP系统接入（SAP/用友/金蝶等），保留模拟数据用于开发/演示\n2. **跨Agent协作流程细化**：细粒度任务协议、状态同步、链路追踪、并行/串行混合执行引擎\n3. **错误处理和状态管理**：统一异常中间件、任务状态机（pending→running→success/failed/retry）、重试策略、操作日志\n\n## 系统架构\n\n```\n┌─────────────────────────────────────────────────────────────┐\n│                     用户请求 (自然语言)                       │\n└───────────────────────┬─────────────────────────────────────┘\n                        │\n                        ▼\n┌─────────────────────────────────────────────────────────────┐\n│              Orchestrator (指挥智能体) v3.0                   │\n│  ┌─────────────┐ ┌─────────────┐ ┌──────────────────────┐  │\n│  │ 意图识别    │→│ 任务拆解    │→│ 智能体调度 (串行/并行) │  │\n│  └─────────────┘ └─────────────┘ └──────────────────────┘  │\n│  ┌──────────────────────────────────────────────────────┐  │\n│  │  ExecutionLayer: EngineRouter（三引擎热切换）           │  │\n│  │  ┌────────────┐ ┌─────────────┐ ┌──────────────────┐  │  │\n│  │  │ClaudeMA    │ │LocalEngine  │ │DeepSeekEngine    │  │  │\n│  │  │(通用任务)  │ │(垂直知识)   │ │(合规场景)        │  │  │\n│  │  └────────────┘ └─────────────┘ └──────────────────┘  │  │\n│  └──────────────────────────────────────────────────────┘  │\n└───────────────────────┬─────────────────────────────────────┘\n                        │\n        ┌───────────────┼───────────────┬───────────────┐\n        ▼               ▼               ▼               ▼\n┌───────────────┐ ┌─────────────┐ ┌───────────────┐ ┌───────────────┐\n│Inventory Agent│ │Logistics    │ │Procurement    │ │Finance Agent  │\n│(库存智能体)   │ │Agent        │ │Agent          │ │(财务智能体)   │\n│               │ │(物流智能体) │ │(采购智能体)   │ │               │\n└───────┬───────┘ └──────┬──────┘ └───────┬───────┘ └───────┬───────┘\n        │                 │                 │                 │\n        ▼                 ▼                 ▼                 ▼\n┌─────────────────────────────────────────────────────────────┐\n│          API Integration Layer（v2.0）                        │\n│  ┌──────────────┐  ┌──────────────┐  ┌──────────────────┐   │\n│  │ SAP适配器    │  │ 用友适配器   │  │ 健康监控+断路器  │   │\n│  │ 金蝶适配器   │  │ 通用REST    │  │ 故障自动降级     │   │\n│  └──────────────┘  └──────────────┘  └──────────────────┘   │\n└─────────────────────────────────────────────────────────────┘\n```\n\n## 执行引擎对比\n\n| 维度 | ClaudeMAEngine | LocalEngine | DeepSeekEngine |\n|------|---------------|-------------|----------------|\n| 场景 | 通用开发任务 | 垂直行业任务 | 合规场景 |\n| 凭证管理 | ✅ 官方托管 | ✅ 完全本地 | ✅ 需配置 |\n| 自进化 | ❌ | ✅ M-A3独有 | ❌ |\n| 垂直知识 | ❌ 需自建 | ✅ 塑化行业 | ❌ |\n| 合规认证 | ❌ | ✅ 完全离线 | ✅ 国产合规 |\n| 流式输出 | ✅ | ✅ | ✅ |\n| 多模态 | ✅ | ❌ | ❌ |\n\n## 目录结构\n\n```\nagent-cluster/\n├── __init__.py                   # 包入口（v3.0）\n│\n├── orchestrator.py               # 指挥智能体（核心调度器）v3.0 新增 execute_with_engine\n├── README.md                     # 本文件\n│\n├── execution/                    # 【新增 v3.0】执行引擎抽象层\n│   ├── __init__.py              # 包入口，统一导出\n│   ├── engine_base.py          # ExecutionEngine 抽象基类 + ExecutionResult\n│   ├── engine_router.py         # EngineRouter 路由器 + RoutingRule/RoutingContext\n│   ├── claude_ma_engine.py     # Claude Managed Agents 适配器\n│   ├── local_engine.py         # 本地自建引擎（Orchestrator 原有逻辑迁移）\n│   └── deepseek_engine.py      # 国产模型适配器（DeepSeek/华为等）\n│\n├── api_integration/              # 【v2.0】真实API接入层\n│   ├── __init__.py\n│   ├── api_adapter.py           # 多ERP适配器（SAP/用友/金蝶）\n│   ├── api_config.py            # 配置化管理（环境变量/YAML）\n│   ├── api_health.py            # 健康检查+断路器+告警\n│   └── mock_data.py            # 模拟数据（开发/演示模式）\n│\n├── collaboration/                # 【新增】跨Agent协作流程\n│   ├── __init__.py\n│   ├── task_protocol.py         # 细粒度任务协议（TaskMessage）\n│   ├── state_sync.py            # Agent间状态同步+TTL+订阅通知\n│   ├── trace_tracker.py         # 协作链路追踪+Mermaid时序图\n│   └── workflow_engine.py        # 混合执行引擎（串行/并行/混合）\n│\n├── error_handling/              # 【新增】错误处理与状态管理\n│   ├── __init__.py\n│   ├── task_state_machine.py    # 任务状态机（7种状态+转换规则）\n│   ├── exception_middleware.py  # 统一异常处理+分类+告警\n│   ├── retry_policy.py          # 重试策略（5种）+条件重试\n│   └── operation_log.py        # 操作日志+脱敏+合规报告\n│\n├── specialists/                 # 专业智能体\n│   ├── inventory_agent.py      # 库存智能体\n│   ├── logistics_agent.py       # 物流智能体\n│   ├── procurement_agent.py     # 采购智能体\n│   ├── finance_agent.py         # 财务智能体\n│   └── doc_agent.py            # 工艺文档智能体\n│\n├── mcp_servers/                 # MCP协议封装\n│   ├── erp_server.py           # ERP系统接口\n│   ├── wms_server.py           # WMS仓库管理接口\n│   └── srm_server.py           # SRM供应商管理接口\n│\n├── safety/                      # 安全围栏\n│   ├── permission_manager.py    # RBAC权限管理\n│   ├── audit_logger.py          # 全链路审计日志\n│   └── human_loop.py            # 人机回环审批\n│\n├── config/                     # 配置文件\n│   ├── agents.yaml             # 智能体定义\n│   ├── workflows.yaml          # 工作流配置\n│   ├── permissions.yaml        # 权限矩阵\n│   └── engines.yaml            # 【新增 v3.0】引擎路由规则配置\n│\n└── tests/                     # 【新增 v3.0】单元测试\n    └── test_execution_engines.py  # 执行引擎测试套件\n```\n\n## 快速开始\n\n### 环境要求\n\n- Python 3.10+\n- 依赖包：\n  ```bash\n  pip install pyyaml fastapi uvicorn httpx aiofiles\n  ```\n\n### 运行演示\n\n```bash\n# 完整演示（指挥智能体）\ncd agent-cluster\npython orchestrator.py\n\n# 单独测试各智能体\npython -m specialists.inventory_agent\npython -m specialists.procurement_agent\npython -m specialists.finance_agent\n```\n\n---\n\n## API接入层详解（v2.0新增）\n\n### 支持的ERP类型\n\n| ERP类型 | 适配器 | API协议 | 配置示例 |\n|---------|--------|---------|---------|\n| SAP S/4HANA | `SAPERPAdapter` | OData/REST | `SAP_BASE_URL` |\n| 用友U8/NC/YonBIP | `YonyouERPAdapter` | REST API | `YONYOU_BASE_URL` |\n| 金蝶K3 Cloud/EAS | `KingdeeERPAdapter`（可扩展） | REST API | `KINGDEE_BASE_URL` |\n| 通用REST | `CustomRESTAdapter`（可扩展） | OpenAPI | `CUSTOM_BASE_URL` |\n| 模拟模式 | `MockDataGenerator` | - | `SYSTEM_MODE=demo` |\n\n### 配置方式\n\n**方式1：环境变量**\n```bash\nexport SYSTEM_MODE=demo          # demo/production/development\nexport SAP_BASE_URL=https://sap.example.com\nexport SAP_API_KEY=sk-xxx\nexport YONYOU_BASE_URL=https://yonyou.example.com\nexport YONYOU_APPKEY=your_appkey\n```\n\n**方式2：YAML配置文件**\n```yaml\nsystem:\n  mode: demo\n  log_level: INFO\n  enable_trace: true\n  demo_variance: 0.1\n\nerp_systems:\n  - name: sap_primary\n    erp_type: sap\n    is_primary: true\n    base_url: https://sap.example.com\n    auth_type: bearer\n    api_key: ${SAP_API_KEY}\n    timeout: 30.0\n    circuit_breaker_threshold: 5\n```\n\n### 代码示例\n\n```python\nfrom api_integration import APIConfigManager, MockDataGenerator\n\n# 自动从环境变量加载配置\nconfig = APIConfigManager.from_env()\n\n# 演示模式使用模拟数据\nif config.system.mode == \"demo\":\n    mock = MockDataGenerator(variance=0.1)\n    response = mock.query_inventory(sku=\"SKU001\")\n    print(response.data)\n```\n\n---\n\n## 协作流程详解（v2.0新增）\n\n### 细粒度任务协议\n\n```python\nfrom collaboration import TaskMessage, TaskContext, TaskPriority, TaskMode\n\n# 创建任务消息\ntask = TaskMessage(\n    agent_name=\"inventory_agent\",\n    action=\"query_stock\",\n    parameters={\"sku\": \"SKU001\", \"warehouse\": \"华东仓\"},\n    priority=TaskPriority.HIGH,\n    mode=TaskMode.SERIAL,\n    timeout_seconds=30.0,\n    max_retries=3,\n    dependency=TaskDependency(\n        depends_on=[\"task_001\", \"task_002\"],  # 依赖的任务ID\n        blocking=True,\n        shared_context=[\"budget_summary\"],     # 需要共享的上下文\n    ),\n    context=TaskContext(request_id=\"REQ001\", trace_id=\"trace_xxx\"),\n)\n```\n\n### 状态同步\n\n```python\nfrom collaboration import SharedStateManager\n\nstate = SharedStateManager(agent_id=\"inventory_agent\")\n\n# 设置状态（TTL=300秒）\nawait state.set(\"inventory:SKU001\", {\"qty\": 450}, ttl_seconds=300)\n\n# 订阅变更\nstate.subscribe(\n    \"inventory:*\",\n    lambda key, value, entry: print(f\"库存更新: {key}={value}\"),\n    subscriber=\"logistics_agent\",\n)\n```\n\n### 链路追踪（Mermaid时序图）\n\n```python\nfrom collaboration import CollaborationTracker\n\ntracker = CollaborationTracker()\ntrace_id = tracker.start_trace(\"REQ001\", user_input=\"查询库存\")\n\n# 执行任务并记录\nspan = tracker.start_span(trace_id, \"inventory_agent:query_stock\", SpanType.AGENT)\n# ... 执行逻辑 ...\ntracker.end_span(span, status=\"ok\")\n\n# 导出Mermaid时序图\nprint(tracker.to_mermaid_sequence(trace_id))\n```\n\n输出示例：\n```mermaid\nsequenceDiagram\n    participant 用户\n    participant inventory_agent\n    participant erp_system\n    用户->>+inventory_agent: query_stock(SKU001)\n    inventory_agent->>+erp_system: GET /api/stock\n    erp_system-->>-inventory_agent: {qty: 450}\n    inventory_agent-->>-用户: {status: ok}\n```\n\n---\n\n## 错误处理详解（v2.0新增）\n\n### 状态机\n\n```python\nfrom error_handling import TaskStateMachine, State\n\nsm = TaskStateMachine(auto_retry=True, default_max_retries=3)\n\n# 创建任务\ntask = sm.create_task(\"task_001\", \"查询库存\", \"inventory_agent\")\n\n# 状态转换（自动校验合法性）\nsm.start(\"task_001\")         # pending → running\nsm.succeed(\"task_001\", result={\"qty\": 450})  # → success\n\n# 失败时自动重试（指数退避）\nsm.fail(\"task_001\", \"网络超时\")\n# → running → retry（第1次，1s后）→ running → ...\n# → running → retry（第3次，8s后）→ running → ...\n# → failed（超过最大重试次数）\n```\n\n### 统一异常处理\n\n```python\nfrom error_handling import ExceptionMiddleware, handle_exceptions\n\nmiddleware = ExceptionMiddleware()\n\ntry:\n    # ERP API调用\n    response = await adapter.query_inventory(sku=\"SKU001\")\nexcept Exception as e:\n    error = middleware.handle(e, source=\"inventory_agent\", request_id=\"REQ001\")\n    print(error.to_dict())\n    # {\n    #   \"error_id\": \"a1b2c3d4\",\n    #   \"category\": \"net\n\nArchive v3.0.2: 144 files, 521512 bytes\n\nFiles: __init__.py (406b), _meta.json (132b), 30-expansion/30-agents-design.md (51910b), 30-expansion/agent_protocol.md (16120b), 30-expansion/agent_registry.yaml (18892b), 30-expansion/IMPLEMENTATION_PLAN.md (13710b), api_integration/__init__.py (936b), api_integration/api_adapter.py (21482b), api_integration/api_config.py (9243b), api_integration/api_health.py (8629b), api_integration/deepseek_integration.md (11899b), api_integration/mock_data.py (14178b), cms_approvals/04cf553c-e0c9-4732-870b-83f96e443b52.json (533b), cms_approvals/1c9e92a7-46f3-4a6d-a0d9-59c60ff0fee1.json (692b), cms_approvals/265648cd-fd95-4b2c-a1e7-4051cbe771cc.json (569b), cms_approvals/2c363c59-d353-4e3d-a2b6-aa4e321fa278.json (569b), cms_approvals/31ff6cde-b148-47d1-980f-6953a7e1fe98.json (569b), cms_approvals/326cf306-f324-478f-9abb-952657fd8a58.json (569b), cms_approvals/380f9c96-b301-4cbc-9a61-d508323241bf.json (717b), cms_approvals/5709cd14-a912-4bb9-a877-e8fd3c029ec2.json (569b), cms_approvals/a2941003-ecb3-41bc-8522-088b9bae1f41.json (701b), cms_approvals/b260ecbf-649d-4da4-b064-3093114d7ca7.json (704b), cms_approvals/d55ab274-327d-4090-985d-8c8c4d795c5c.json (683b), cms_approvals/de294f4f-a55b-43b2-bbae-3840b81f2e85.json (674b), cms_approvals/eb03a9c1-ecb6-4007-ada6-aaefd49ef4ac.json (533b), cms_executor/__init__.py (2412b), cms_executor/agent_integration.py (26443b), cms_executor/ARCHITECTURE.md (14538b), cms_executor/connectors/__init__.py (891b), cms_executor/connectors/amazon_connector.py (14762b), cms_executor/connectors/base_connector.py (27345b), cms_executor/connectors/magento_connector.py (16667b), cms_executor/connectors/shopify_connector.py (16183b), cms_executor/connectors/wordpress_connector.py (11068b), cms_executor/engine/__init__.py (642b), cms_executor/engine/approval.py (11726b), cms_executor/engine/audit.py (17904b), cms_executor/engine/executor.py (19384b), cms_executor/engine/rollback.py (15292b), cms_executor/tests/__init__.py (224b), cms_executor/tests/test_agent_integration.py (14358b), cms_executor/tests/test_approval.py (8408b), cms_executor/tests/test_base_connector.py (11999b), cms_executor/tests/test_connectors.py (14832b), cms_executor/tests/test_executor.py (13274b), cms_executor/tests/test_rollback.py (9323b), cms-executor/connectors/__init__.py (457b), cms-executor/connectors/amazon_connector.py (26421b), cms-executor/connectors/base_connector.py (5652b), cms-executor/connectors/shopify_connector.py (25896b), cms-executor/connectors/wordpress_connector.py (8616b), cms-executor/engine/__init__.py (168b), cms-executor/engine/approval.py (9193b), cms-executor/engine/rollback.py (10602b), cms-executor/README.md (12573b), cms-executor/tests/test_cms_executor.py (13195b), cms-executor/tests/test_connectors.py (31610b), collaboration/__init__.py (1067b), collaboration/state_sync.py (10976b), collaboration/task_protocol.py (15311b), collaboration/tests/__init__.py (22570b), collaboration/trace_tracker.py (11782b), collaboration/workflow_engine.py (12650b), config/agents.yaml (3021b), config/engines.yaml (7322b), config/permissions.yaml (3718b), config/workflows.yaml (3799b), error_handling/__init__.py (1260b), error_handling/exception_middleware.py (11348b), error_handling/operation_log.py (15910b), error_handling/retry_policy.py (10861b), error_handling/task_state_machine.py (13609b), execution/__init__.py (1564b), execution/circuit_breaker.py (40704b), execution/claude_ma_engine.py (8921b), execution/deepseek_engine.py (37869b), execution/engine_base.py (23172b), execution/engine_router.py (37840b), execution/gpt6_engine.py (27838b), execution/local_engine.py (11985b)\n\nArchive v3.0.1: 144 files, 521474 bytes\n\nFiles: __init__.py (406b), _meta.json (132b), 30-expansion/30-agents-design.md (51910b), 30-expansion/agent_protocol.md (16120b), 30-expansion/agent_registry.yaml (18892b), 30-expansion/IMPLEMENTATION_PLAN.md (13710b), api_integration/__init__.py (936b), api_integration/api_adapter.py (21482b), api_integration/api_config.py (9243b), api_integration/api_health.py (8629b), api_integration/deepseek_integration.md (11899b), api_integration/mock_data.py (14178b), cms_approvals/04cf553c-e0c9-4732-870b-83f96e443b52.json (533b), cms_approvals/1c9e92a7-46f3-4a6d-a0d9-59c60ff0fee1.json (692b), cms_approvals/265648cd-fd95-4b2c-a1e7-4051cbe771cc.json (569b), cms_approvals/2c363c59-d353-4e3d-a2b6-aa4e321fa278.json (569b), cms_approvals/31ff6cde-b148-47d1-980f-6953a7e1fe98.json (569b), cms_approvals/326cf306-f324-478f-9abb-952657fd8a58.json (569b), cms_approvals/380f9c96-b301-4cbc-9a61-d508323241bf.json (717b), cms_approvals/5709cd14-a912-4bb9-a877-e8fd3c029ec2.json (569b), cms_approvals/a2941003-ecb3-41bc-8522-088b9bae1f41.json (701b), cms_approvals/b260ecbf-649d-4da4-b064-3093114d7ca7.json (704b), cms_approvals/d55ab274-327d-4090-985d-8c8c4d795c5c.json (683b), cms_approvals/de294f4f-a55b-43b2-bbae-3840b81f2e85.json (674b), cms_approvals/eb03a9c1-ecb6-4007-ada6-aaefd49ef4ac.json (533b), cms_executor/__init__.py (2412b), cms_executor/agent_integration.py (26443b), cms_executor/ARCHITECTURE.md (14538b), cms_executor/connectors/__init__.py (891b), cms_executor/connectors/amazon_connector.py (14762b), cms_executor/connectors/base_connector.py (27345b), cms_executor/connectors/magento_connector.py (16667b), cms_executor/connectors/shopify_connector.py (16183b), cms_executor/connectors/wordpress_connector.py (11068b), cms_executor/engine/__init__.py (642b), cms_executor/engine/approval.py (11726b), cms_executor/engine/audit.py (17904b), cms_executor/engine/executor.py (19384b), cms_executor/engine/rollback.py (15292b), cms_executor/tests/__init__.py (224b), cms_executor/tests/test_agent_integration.py (14358b), cms_executor/tests/test_approval.py (8408b), cms_executor/tests/test_base_connector.py (11999b), cms_executor/tests/test_connectors.py (14832b), cms_executor/tests/test_executor.py (13274b), cms_executor/tests/test_rollback.py (9323b), cms-executor/connectors/__init__.py (457b), cms-executor/connectors/amazon_connector.py (26421b), cms-executor/connectors/base_connector.py (5652b), cms-executor/connectors/shopify_connector.py (25896b), cms-executor/connectors/wordpress_connector.py (8616b), cms-executor/engine/__init__.py (168b), cms-executor/engine/approval.py (9193b), cms-executor/engine/rollback.py (10602b), cms-executor/README.md (12573b), cms-executor/tests/test_cms_executor.py (13195b), cms-executor/tests/test_connectors.py (31610b), collaboration/__init__.py (1067b), collaboration/state_sync.py (10976b), collaboration/task_protocol.py (15311b), collaboration/tests/__init__.py (22570b), collaboration/trace_tracker.py (11782b), collaboration/workflow_engine.py (12650b), config/agents.yaml (3021b), config/engines.yaml (7322b), config/permissions.yaml (3718b), config/workflows.yaml (3799b), error_handling/__init__.py (1260b), error_handling/exception_middleware.py (11348b), error_handling/operation_log.py (15910b), error_handling/retry_policy.py (10861b), error_handling/task_state_machine.py (13609b), execution/__init__.py (1564b), execution/circuit_breaker.py (40704b), execution/claude_ma_engine.py (8921b), execution/deepseek_engine.py (37869b), execution/engine_base.py (23172b), execution/engine_router.py (37840b), execution/gpt6_engine.py (27838b), execution/local_engine.py (11985b)\n\nArchive v1.0.0: 144 files, 521451 bytes\n\nFiles: __init__.py (406b), 30-expansion/30-agents-design.md (51910b), 30-expansion/agent_protocol.md (16120b), 30-expansion/agent_registry.yaml (18892b), 30-expansion/IMPLEMENTATION_PLAN.md (13710b), api_integration/__init__.py (936b), api_integration/api_adapter.py (21482b), api_integration/api_config.py (9243b), api_integration/api_health.py (8629b), api_integration/deepseek_integration.md (11899b), api_integration/mock_data.py (14178b), cms_approvals/04cf553c-e0c9-4732-870b-83f96e443b52.json (533b), cms_approvals/1c9e92a7-46f3-4a6d-a0d9-59c60ff0fee1.json (692b), cms_approvals/265648cd-fd95-4b2c-a1e7-4051cbe771cc.json (569b), cms_approvals/2c363c59-d353-4e3d-a2b6-aa4e321fa278.json (569b), cms_approvals/31ff6cde-b148-47d1-980f-6953a7e1fe98.json (569b), cms_approvals/326cf306-f324-478f-9abb-952657fd8a58.json (569b), cms_approvals/380f9c96-b301-4cbc-9a61-d508323241bf.json (717b), cms_approvals/5709cd14-a912-4bb9-a877-e8fd3c029ec2.json (569b), cms_approvals/a2941003-ecb3-41bc-8522-088b9bae1f41.json (701b), cms_approvals/b260ecbf-649d-4da4-b064-3093114d7ca7.json (704b), cms_approvals/d55ab274-327d-4090-985d-8c8c4d795c5c.json (683b), cms_approvals/de294f4f-a55b-43b2-bbae-3840b81f2e85.json (674b), cms_approvals/eb03a9c1-ecb6-4007-ada6-aaefd49ef4ac.json (533b), cms_executor/__init__.py (2412b), cms_executor/agent_integration.py (26443b), cms_executor/ARCHITECTURE.md (14538b), cms_executor/connectors/__init__.py (891b), cms_executor/connectors/amazon_connector.py (14762b), cms_executor/connectors/base_connector.py (27345b), cms_executor/connectors/magento_connector.py (16667b), cms_executor/connectors/shopify_connector.py (16183b), cms_executor/connectors/wordpress_connector.py (11068b), cms_executor/engine/__init__.py (642b), cms_executor/engine/approval.py (11726b), cms_executor/engine/audit.py (17904b), cms_executor/engine/executor.py (19384b), cms_executor/engine/rollback.py (15292b), cms_executor/tests/__init__.py (224b), cms_executor/tests/test_agent_integration.py (14358b), cms_executor/tests/test_approval.py (8408b), cms_executor/tests/test_base_connector.py (11999b), cms_executor/tests/test_connectors.py (14832b), cms_executor/tests/test_executor.py (13274b), cms_executor/tests/test_rollback.py (9323b), cms-executor/connectors/__init__.py (457b), cms-executor/connectors/amazon_connector.py (26421b), cms-executor/connectors/base_connector.py (5652b), cms-executor/connectors/shopify_connector.py (25896b), cms-executor/connectors/wordpress_connector.py (8616b), cms-executor/engine/__init__.py (168b), cms-executor/engine/approval.py (9193b), cms-executor/engine/rollback.py (10602b), cms-executor/README.md (12573b), cms-executor/tests/test_cms_executor.py (13195b), cms-executor/tests/test_connectors.py (31610b), collaboration/__init__.py (1067b), collaboration/state_sync.py (10976b), collaboration/task_protocol.py (15311b), collaboration/tests/__init__.py (22570b), collaboration/trace_tracker.py (11782b), collaboration/workflow_engine.py (12650b), config/agents.yaml (3021b), config/engines.yaml (7322b), config/permissions.yaml (3718b), config/workflows.yaml (3799b), error_handling/__init__.py (1260b), error_handling/exception_middleware.py (11348b), error_handling/o...","readmeExcerpt":"Skill: Agent Cluster Owner: wangm-a3 Summary: 亚马逊外贸B2B多CMS Agent协作系统。支持Shopify/WooCommerce/Magento三大平台，A2A架构协调库存/采购/财务/物流四大专家Agent，三层安全网保障审批合规。即装即用，零配置开箱。 Tags: latest:3.0.4 Version history: v3.0.4 | 2026-05-01T15:03:39.752Z | user Security fixes: removed eval/exec patterns, replaced exposed API key placeholders, cleaned prompt patterns v3.0.3 | 2026-04-29T03:50:57.891Z | user Remove hardcoded key patterns from test ","codeSnippets":[],"executableExamples":[{"language":"text","snippet":"用户请求 → Orchestrator（意图识别→任务拆解→智能体调度）\n    ↓\n20个专业Agent：采购/生产/销售/财务/运营/战略/研发/人力/合规\n    ↓\nAPI Integration Layer（v2.0新增）\n  ├─ SAP/用友/金蝶适配器（真实ERP）\n  └─ 断路器+健康检查+Mock降级"},{"language":"text","snippet":"agent-cluster/\n├── orchestrator.py              # 指挥智能体（核心调度器）\n├── api_integration/              # v2.0新增：真实API接入层\n│   ├── api_adapter.py           # 多ERP适配器（SAP/用友/金蝶）\n│   ├── api_config.py            # 配置化管理\n│   ├── api_health.py            # 健康检查+断路器\n│   └── mock_data.py             # 模拟数据（开发/演示）\n├── collaboration/                # v2.0新增：跨Agent协作\n│   ├── task_protocol.py         # 细粒度任务协议\n│   ├── state_sync.py            # 状态同步+TTL+pub/sub\n│   ├── trace_tracker.py         # 全链路追踪+Mermaid\n│   └── workflow_engine.py       # 混合执行引擎\n├── error_handling/              # v2.0新增：错误处理\n│   ├── task_state_machine.py    # 7状态任务状态机\n│   ├── exception_middleware.py  # 统一异常处理\n│   ├── retry_policy.py          # 5种重试策略\n│   └── operation_log.py         # 操作日志+脱敏+合规\n├── specialists/                 # 专业智能体\n│   ├── inventory_agent.py      # 库存智能体\n│   ├── logistics_agent.py       # 物流智能体\n│   ├── procurement_agent.py     # 采购智能体\n│   ├── finance_agent.py         # 财务智能体\n│   └── doc_agent.py            # 工艺文档智能体\n├── mcp_servers/                 # MCP协议封装\n│   ├── erp_server.py           # ERP系统接口\n│   ├── wms_server.py           # WMS仓库管理接口\n│   └── srm_server.py           # SRM供应商管理接口\n├── safety/                      # 安全围栏\n│   ├── permission_manager.py    # RBAC权限管理\n│   ├── audit_logger.py          # 全链路审计日志\n│   └── human_loop.py            # 人机回环审批\n└── config/                      # 配置文件\n    ├── agents.yaml             # 智能体定义\n    ├── workflows.yaml          # 工作流配置\n    └── permissions.yaml        # 权限矩阵"},{"language":"bash","snippet":"cd agent-cluster\npython orchestrator.py"},{"language":"bash","snippet":"export SYSTEM_MODE=production\nexport SAP_BASE_URL=https://sap.example.com\nexport SAP_API_KEY=sk-xxx\nexport YONYOU_BASE_URL=https://yonyou.example.com"},{"language":"text","snippet":"cms-executor/\n├── connectors/\n│   ├── __init__.py              # 统一导出\n│   ├── base_connector.py        # 抽象基类（所有 CMS 平台通用接口）\n│   ├── wordpress_connector.py   # WordPress REST API 实现\n│   ├── shopify_connector.py     # Shopify GraphQL Admin API 实现\n│   └── amazon_connector.py      # Amazon SP-API 实现\n├── engine/\n│   ├── __init__.py\n│   ├── approval.py              # 多级审批引擎\n│   └── rollback.py              # 变更追踪与回滚\n└── tests/\n    ├── test_cms_executor.py     # WordPress & 审批引擎测试\n    └── test_connectors.py       # Shopify & Amazon 连接器测试"},{"language":"python","snippet":"from connectors import ShopifyConnector, CMSCredential, ContentPayload\n\n# 初始化连接器\ncred = CMSCredential(\n    url=\"https://your-shop.myshopify.com\",\n    api_key=\"<YOUR_SHOPIFY_ACCESS_TOKEN>\",   # 访问令牌\n)\nshopify = ShopifyConnector(cred)\n\n# 验证连接\nif shopify.authenticate():\n    print(\"✅ Shopify 连接成功\")\n\n# ── 商品 CRUD ────────────────────────────────────────\nfrom connectors.shopify_connector import ProductPayload, VariantPayload\n\n# 方式一：使用 ContentPayload（兼容基类）\npayload = ContentPayload(title=\"My Product\", content=\"<p>Description</p>\", status=\"draft\")\nresult = shopify.create_content(payload)\nprint(f\"✅ 商品已创建，ID={result['id']}, handle={result['handle']}\")\n\n# 方式二：使用 ProductPayload（Shopify 原生字段）\nfrom connectors.shopify_connector import ProductPayload, VariantPayload, ImagePayload, SEOPayload\nproduct = ProductPayload(\n    title=\"Premium Widget\",\n    body_html=\"<p>High quality widget</p>\",\n    vendor=\"Acme Corp\",\n    product_type=\"Electronics\",\n    status=\"active\",\n    tags=[\"widget\", \"premium\"],\n    variants=[\n        VariantPayload(\n            title=\"Blue / Large\",\n            price=\"29.99\",\n            sku=\"WGT-BL-L\",\n            inventory_quantity=100,\n            option1=\"Blue\", option2=\"Large\",\n        ),\n        VariantPayload(\n            title=\"Red / Large\",\n            price=\"29.99\",\n            sku=\"WGT-RE-L\",\n            inventory_quantity=50,\n            option1=\"Red\", option2=\"Large\",\n        ),\n    ],\n    images=[\n        ImagePayload(src=\"https://cdn.example.com/widget.jpg\", alt_text=\"Widget front view\"),\n    ],\n    seo=SEOPayload(title=\"Premium Widget - Acme Corp\", description=\"Shop premium widgets\"),\n)\n# 通过内部 GraphQL API 直接创建\nresp = shopify._graphql(shopify._build_product_create_mutation(product))\n\n# ── 变体操作 ──────────────────────────────────────────\nvariant = VariantPayload(\n    title=\"Green / Medium\",\n    price=\"24.99\",\n    sku=\"WGT-GN-M\",\n    inventory_quantity=30,\n)\nshopify.create_variant(product_id=123, variant=variant)\nshopify.update_variant(variant_id=555, va"}],"parameters":null,"dependencies":[],"permissions":[],"extractedFiles":[{"path":"SKILL.md","content":"---\nname: foreign-trade-silicon-army\ndescription: 亚马逊外贸B2B多CMS Agent协作系统。支持Shopify/WooCommerce/Magento三大平台，A2A架构协调库存/采购/财务/物流四大专家Agent，三层安全网保障审批合规。即装即用，零配置开箱。\nversion: 3.0.4\ntriggers:\n  - \"外贸\"\n  - \"CMS\"\n  - \"Shopify\"\n  - \"WooCommerce\"\n  - \"Magento\"\n  - \"库存\"\n  - \"采购\"\n  - \"Amazon\"\n  - \"B2B\"\ntags:\n  - agent\n  - a2a\n  - cms\n  - ecommerce\n  - b2b\n  - foreign-trade\n  - multi-agent\nrequires:\n  python_packages:\n    - httpx\n    - pyyaml\n    - fastapi\n    - uvicorn\n  env:\n    - SYSTEM_MODE\n    - SAP_BASE_URL\n    - SAP_API_KEY\n    - YONYOU_BASE_URL\n    - YONYOU_APPKEY\n    - KINGDEE_BASE_URL\n    - ANTHROPIC_API_KEY\n    - DEEPSEEK_API_KEY\nhomepage: https://github.com/WangM-A3/agent-cluster\nauthors:\n  - WangM-A3\npricing: free,pro=9.9,enterprise\n---\n\n# 产业互联网硅基军团 v2.0\n\n> 企业级Multi-Agent智能体集群系统，基于**1+N架构**（1个幕僚长+20个专业Agent），参考OpenClaw Main Agent、腾讯ADP Router设计。v2.0全面升级：真实ERP API接入、协作流程细化、错误处理体系。\n\n## 核心能力\n\n### v2.0三大升级\n\n**1. 真实API接入层（api_integration/）**\n- 多ERP适配器：SAP S/4HANA、用友U8/NC/YonBIP、金蝶K3 Cloud/EAS、通用REST\n- 断路器模式（Circuit Breaker）+ 故障自动降级\n- 健康检查轮询（10s间隔，自动摘除异常节点）\n- 演示模式保留（MockDataGenerator，variance=0.1随机波动）\n\n**2. 跨Agent协作细化（collaboration/）**\n- 细粒度任务协议（TaskMessage）：依赖声明、优先级、TTL\n- 状态同步（SharedStateManager）：TTL+pub/sub通知机制\n- 全链路追踪（CollaborationTracker）：trace_id/span_id + Mermaid时序图可视化\n\n**3. 错误处理与状态管理（error_handling/）**\n- 7状态任务状态机：pending→running→success/failed/retry/timeout/cancelled\n- 10类异常自动分类：VALIDATION/NETWORK/TIMEOUT/AUTH/RESOURCE/NOT_FOUND等\n- 5种重试策略：FIXED/EXPONENTIAL/FIBONACCI/JITTER/ADAPTIVE\n- 敏感信息脱敏 + SOC2合规审计日志\n\n## 系统架构\n\n```\n用户请求 → Orchestrator（意图识别→任务拆解→智能体调度）\n    ↓\n20个专业Agent：采购/生产/销售/财务/运营/战略/研发/人力/合规\n    ↓\nAPI Integration Layer（v2.0新增）\n  ├─ SAP/用友/金蝶适配器（真实ERP）\n  └─ 断路器+健康检查+Mock降级\n```\n\n## 目录结构\n\n```\nagent-cluster/\n├── orchestrator.py              # 指挥智能体（核心调度器）\n├── api_integration/              # v2.0新增：真实API接入层\n│   ├── api_adapter.py           # 多ERP适配器（SAP/用友/金蝶）\n│   ├── api_config.py            # 配置化管理\n│   ├── api_health.py            # 健康检查+断路器\n│   └── mock_data.py             # 模拟数据（开发/演示）\n├── collaboration/                # v2.0新增：跨Agent协作\n│   ├── task_protocol.py         # 细粒度任务协议\n│   ├── state_sync.py            # 状态同步+TTL+pub/sub\n│   ├── trace_tracker.py         # 全链路追踪+Mermaid\n│   └── workflow_engine.py       # 混合执行引擎\n├── error_handling/              # v2.0新增：错误处理\n│   ├── task_state_machine.py    # 7状态任务状态机\n│   ├── exception_middleware.py  # 统一异常处理\n│   ├── retry_policy.py          # 5种重试策略\n│   └── operation_log.py         # 操作日志+脱敏+合规\n├── specialists/                 # 专业智能体\n│   ├── inventory_agent.py      # 库存智能体\n│   ├── logistics_agent.py       # 物流智能体\n│   ├── procurement_agent.py     # 采购智能体\n│   ├── finance_agent.py         # 财务智能体\n│   └── doc_agent.py            # 工艺文档智能体\n├── mcp_servers/                 # MCP协议封装\n│   ├── erp_server.py           # ERP系统接口\n│   ├── wms_server.py           # WMS仓库管理接口\n│   └── srm_server.py           # SRM供应商管理接口\n├── safety/                      # 安全围栏\n│   ├── permission_manager.py    # RBAC权限管理\n│   ├── audit"},{"path":"cms-executor/README.md","content":"# CMS Executor - 多平台 CMS 连接器\n\n## 概述\n\n本模块提供多平台（WordPress / Shopify / Amazon）CMS 直连执行能力，支持通过平台原生 API 创建/更新/删除内容，配合审批引擎和回滚机制实现安全的内容管理。\n\n## 目录结构\n\n```\ncms-executor/\n├── connectors/\n│   ├── __init__.py              # 统一导出\n│   ├── base_connector.py        # 抽象基类（所有 CMS 平台通用接口）\n│   ├── wordpress_connector.py   # WordPress REST API 实现\n│   ├── shopify_connector.py     # Shopify GraphQL Admin API 实现\n│   └── amazon_connector.py      # Amazon SP-API 实现\n├── engine/\n│   ├── __init__.py\n│   ├── approval.py              # 多级审批引擎\n│   └── rollback.py              # 变更追踪与回滚\n└── tests/\n    ├── test_cms_executor.py     # WordPress & 审批引擎测试\n    └── test_connectors.py       # Shopify & Amazon 连接器测试\n```\n\n---\n\n## 平台支持\n\n| 平台 | API | 认证方式 |\n|------|-----|---------|\n| WordPress | REST API v2 | Application Passwords |\n| Shopify | GraphQL Admin API 2024-01 | OAuth / API Key |\n| Amazon | SP-API | LWA OAuth |\n\n---\n\n## 1. Shopify 连接器\n\n### 认证配置\n\n1. 在 Shopify 后台创建私有应用或自定义应用\n2. 配置 API 权限（Products, Variants, Images）\n3. 获取访问令牌（Access Token）\n\n### 基本使用\n\n```python\nfrom connectors import ShopifyConnector, CMSCredential, ContentPayload\n\n# 初始化连接器\ncred = CMSCredential(\n    url=\"https://your-shop.myshopify.com\",\n    api_key=\"<YOUR_SHOPIFY_ACCESS_TOKEN>\",   # 访问令牌\n)\nshopify = ShopifyConnector(cred)\n\n# 验证连接\nif shopify.authenticate():\n    print(\"✅ Shopify 连接成功\")\n\n# ── 商品 CRUD ────────────────────────────────────────\nfrom connectors.shopify_connector import ProductPayload, VariantPayload\n\n# 方式一：使用 ContentPayload（兼容基类）\npayload = ContentPayload(title=\"My Product\", content=\"<p>Description</p>\", status=\"draft\")\nresult = shopify.create_content(payload)\nprint(f\"✅ 商品已创建，ID={result['id']}, handle={result['handle']}\")\n\n# 方式二：使用 ProductPayload（Shopify 原生字段）\nfrom connectors.shopify_connector import ProductPayload, VariantPayload, ImagePayload, SEOPayload\nproduct = ProductPayload(\n    title=\"Premium Widget\",\n    body_html=\"<p>High quality widget</p>\",\n    vendor=\"Acme Corp\",\n    product_type=\"Electronics\",\n    status=\"active\",\n    tags=[\"widget\", \"premium\"],\n    variants=[\n        VariantPayload(\n            title=\"Blue / Large\",\n            price=\"29.99\",\n            sku=\"WGT-BL-L\",\n            inventory_quantity=100,\n            option1=\"Blue\", option2=\"Large\",\n        ),\n        VariantPayload(\n            title=\"Red / Large\",\n            price=\"29.99\",\n            sku=\"WGT-RE-L\",\n            inventory_quantity=50,\n            option1=\"Red\", option2=\"Large\",\n        ),\n    ],\n    images=[\n        ImagePayload(src=\"https://cdn.example.com/widget.jpg\", alt_text=\"Widget front view\"),\n    ],\n    seo=SEOPayload(title=\"Premium Widget - Acme Corp\", description=\"Shop premium widgets\"),\n)\n# 通过内部 GraphQL API 直接创建\nresp = shopify._graphql(shopify._build_product_create_mutation(product))\n\n# ── 变体操作 ──────────────────────────────────────────\nvariant = VariantPayload(\n    title=\"Green / Medium\",\n    price=\"24.99\",\n    sku=\"WGT-GN-M\",\n    inventory_quantity=30,\n)\nshopify.create_variant(product_id=123, variant=variant)\nsh"},{"path":"README.md","content":"# 企业级智能体集群系统 v3.0\n\n> 基于 **1+N** 架构的智能体协作系统，参考 OpenClaw Main Agent、腾讯ADP Router、智己汽车研发设计集群\n\n## 核心升级（v3.0）\n\n本次更新将执行层解耦，支持**三引擎热切换**：\n\n1. **执行层抽象（ExecutionEngine）**：统一接口抽象，支持 Claude MA / Local / DeepSeek 三种引擎\n2. **EngineRouter 智能路由**：基于意图/场景/关键词的自动引擎选择，支持降级\n3. **向后兼容**：现有 `handle_request` API 完全不受影响，新增 `execute_with_engine` API\n\n## 核心升级（v2.0）\n\n本次更新根据用户评测反馈进行了三大改进：\n\n1. **真实API接入层**：适配器模式支持多ERP系统接入（SAP/用友/金蝶等），保留模拟数据用于开发/演示\n2. **跨Agent协作流程细化**：细粒度任务协议、状态同步、链路追踪、并行/串行混合执行引擎\n3. **错误处理和状态管理**：统一异常中间件、任务状态机（pending→running→success/failed/retry）、重试策略、操作日志\n\n## 系统架构\n\n```\n┌─────────────────────────────────────────────────────────────┐\n│                     用户请求 (自然语言)                       │\n└───────────────────────┬─────────────────────────────────────┘\n                        │\n                        ▼\n┌─────────────────────────────────────────────────────────────┐\n│              Orchestrator (指挥智能体) v3.0                   │\n│  ┌─────────────┐ ┌─────────────┐ ┌──────────────────────┐  │\n│  │ 意图识别    │→│ 任务拆解    │→│ 智能体调度 (串行/并行) │  │\n│  └─────────────┘ └─────────────┘ └──────────────────────┘  │\n│  ┌──────────────────────────────────────────────────────┐  │\n│  │  ExecutionLayer: EngineRouter（三引擎热切换）           │  │\n│  │  ┌────────────┐ ┌─────────────┐ ┌──────────────────┐  │  │\n│  │  │ClaudeMA    │ │LocalEngine  │ │DeepSeekEngine    │  │  │\n│  │  │(通用任务)  │ │(垂直知识)   │ │(合规场景)        │  │  │\n│  │  └────────────┘ └─────────────┘ └──────────────────┘  │  │\n│  └──────────────────────────────────────────────────────┘  │\n└───────────────────────┬─────────────────────────────────────┘\n                        │\n        ┌───────────────┼───────────────┬───────────────┐\n        ▼               ▼               ▼               ▼\n┌───────────────┐ ┌─────────────┐ ┌───────────────┐ ┌───────────────┐\n│Inventory Agent│ │Logistics    │ │Procurement    │ │Finance Agent  │\n│(库存智能体)   │ │Agent        │ │Agent          │ │(财务智能体)   │\n│               │ │(物流智能体) │ │(采购智能体)   │ │               │\n└───────┬───────┘ └──────┬──────┘ └───────┬───────┘ └───────┬───────┘\n        │                 │                 │                 │\n        ▼                 ▼                 ▼                 ▼\n┌─────────────────────────────────────────────────────────────┐\n│          API Integration Layer（v2.0）                        │\n│  ┌──────────────┐  ┌──────────────┐  ┌──────────────────┐   │\n│  │ SAP适配器    │  │ 用友适配器   │  │ 健康监控+断路器  │   │\n│  │ 金蝶适配器   │  │ 通用REST    │  │ 故障自动降级     │   │\n│  └──────────────┘  └──────────────┘  └──────────────────┘   │\n└─────────────────────────────────────────────────────────────┘\n```\n\n## 执行引擎对比\n\n| 维度 | ClaudeMAEngine | LocalEngine | DeepSeekEngine |\n|------|---------------|-------------|----------------|\n| 场景 | 通用开发任务 | 垂直行业任务 | 合规场景 |\n| 凭证管理 | ✅ 官方托管 | ✅ 完全本地 | ✅ 需配置 |\n| 自进化 | ❌ | ✅ M-A3独有 | ❌ |\n| 垂直知识 | ❌ 需自建 | ✅ 塑化行业 | ❌ |\n| 合规认证 | ❌ | ✅ 完全离线 | ✅ 国产合规 |\n| 流式输出 | ✅ | ✅ | ✅ |\n| 多模态 | ✅ | ❌ | ❌ |\n\n## 目录结构\n\n```\nagent-cluster/\n├── __init__.py                   "},{"path":"_meta.json","content":"{\n  \"ownerId\": \"kn7ch74w4kf43pffbq4pxda0w584hy2v\",\n  \"slug\": \"agent-cluster\",\n  \"version\": \"3.0.4\",\n  \"publishedAt\": 1777647819752\n}"},{"path":"30-expansion/30-agents-design.md","content":"# M-A3 Agent集群扩展方案：从6个到30个\n> 版本：v1.0 | 日期：2026-04-14 | 作者：M-A3 幕僚长\n\n---\n\n## 一、项目概述\n\n### 1.1 背景与目标\n\n当前 M-A3 集群已实现：\n- **geo-ops-agents**：6个Agent三层架构（市场研究→内容策略→效果监测）\n- **amazon-ops-agents**：多个运营Agent并行协作\n- **硅基军团**：20个产业Agent（库存/物流/采购/财务/生产/销售）\n\n**扩展目标**：\n> 构建全球最大的垂直领域Agent集群（30个专业Agent），覆盖**出海GEO营销**、**亚马逊全链路运营**、**平台基础设施支撑**三大功能域，形成代差级竞争优势。\n\n### 1.2 规模对比\n\n| 维度 | 现有水平 | 扩展后 | 提升倍数 |\n|------|---------|--------|---------|\n| Agent数量 | 6 | 30 | 5× |\n| 功能域 | 1 | 3 | 3域 |\n| 协作模式 | 串行为主 | 并行+串行混合 | 质变 |\n| 覆盖阶段 | 单点 | 全链路 | 全链路 |\n\n### 1.3 三层架构（扩展版）\n\n```\n┌─────────────────────────────────────────────────────────────────┐\n│                    幕僚长 M-A3（Chief of Staff）                    │\n│  意图识别 → 任务拆解 → 智能调度（并行/串行）→ 结果聚合 → 报告生成      │\n└───────────────────────────────┬─────────────────────────────────┘\n                                │ 并行/串行混合调度\n        ┌───────────────────────┼───────────────────────┐\n        ▼                       ▼                       ▼\n┌───────────────┐      ┌───────────────┐      ┌───────────────┐\n│   GEO域       │      │  亚马逊域     │      │  支撑域        │\n│  (10 Agents)  │      │  (10 Agents)  │      │  (10 Agents)   │\n│   第2层       │      │   第2层        │      │   第2层        │\n└───────────────┘      └───────────────┘      └───────────────┘\n        │                       │                       │\n        ▼                       ▼                       ▼\n┌─────────────────────────────────────────────────────────────────┐\n│               外部工具层（第1层：Tool/Plugin）                      │\n│  搜索引擎 / 平台API / 浏览器 / 数据库 / 文件系统 / 邮件系统           │\n└─────────────────────────────────────────────────────────────────┘\n```\n\n---\n\n## 二、Agent详细设计\n\n### 2.1 GEO域（10个）— 出海营销与AI搜索优化\n\n#### 🔍 GEO-01 市场研究Agent\n```\n职责：输入产品/品类 → 输出目标市场机会分析报告\n核心能力：\n  - 全球主要市场（北美/欧盟/东南亚/拉美/中东）容量估算\n  - 目标客群画像生成（年龄/收入/购买习惯/文化偏好）\n  - 市场规模计算（TAM/SAM/SOM三层模型）\n  - 季节性需求分析\n  - 监管政策扫描（GDPR/REACH/FCC等合规要求）\n技能包：geo-operations-assistant, search_web, 数据分析\n输入：产品描述、目标市场列表\n输出：市场机会报告（JSON + 可视化建议）\n关键词触发：[\"市场研究\", \"市场规模\", \"目标市场\", \"TAM\", \"市场机会\"]\n优先级：P0（入口节点）\n```\n\n#### 📊 GEO-02 竞品分析Agent\n```\n职责：深度分析竞争对手的GEO策略，找出攻防机会\n核心能力：\n  - 竞品识别（LLM Brand Detection + 搜索结果验证）\n  - GEO三维度评分：内容权威性×技术SEO×品牌信号\n  - 竞品内容策略拆解（话题/格式/渠道/频率）\n  - 竞品反向链接分析\n  - AI搜索引用率对比（Perplexity/SearchGPT/DeepSearch）\n  - Gap分析：竞品覆盖但我方空白的GEO机会点\n技能包：search_web, 数据分析, 知识图谱\n输入：竞争对手列表/产品关键词\n输出：竞品GEO画像 + 攻防策略建议\n关键词触发：[\"竞品分析\", \"竞争对手\", \"竞品策略\", \"对标分析\"]\n优先级：P1\n```\n\n#### ✍️ GEO-03 内容策略Agent\n```\n职责：制定全渠道内容日历和内容策略\n核心能力：\n  - 话题发现（高频问题/长尾疑问/Trending话题）\n  - 内容形式规划（博客/白皮书/视频脚本/社交帖）\n  - 渠道分布策略（官网/知乎/CSDN/Medium/LinkedIn）\n  - 发布频率优化（基于竞品数据+搜索信号）\n  - 内容复用矩阵（一篇长文→多条短帖→多语言版本）\n  - E-E-A-T 信号植入策略\n技能包：内容生成, 翻译, search_web\n输入：产品信息、目标关键词、渠道偏好\n输出：季度内容日历（CSV）+ 话题优先级排序\n关键词触发：[\"内容策略\", \"内容规划\", \"选题\", \"内容日历\"]\n优先级：P0（核心输出节点）\n```\n\n#### 🌐 GEO-04 多语言优化Agent\n```\n职责：实现GEO内容的全球化与本地化\n核心能力：\n  - 语言市场优先级排序（基于搜索量+购买力）\n  - 本地化关键词研究（文化差异+本地搜索引擎差异）\n  - 地道表达生成（避免机翻感）\n  - 文化适配（节日/习俗/禁忌词）\n  - hreflang标签策略\n  - 本地化内容质量评估（Native Speaker风格评分）\n技能包：翻译, search_web, 内容生成\n输入：原语言内容、目标语言列表\n输出：本地化内容 + hrefl"}],"languages":[],"docsSourceLabel":"CLAWHUB","editorialOverview":"亚马逊外贸B2B多CMS Agent协作系统。支持Shopify/WooCommerce/Magento三大平台，A2A架构协调库存/采购/财务/物流四大专家Agent，三层安全网保障审批合规。即装即用，零配置开箱。 Skill: Agent Cluster Owner: wangm-a3 Summary: 亚马逊外贸B2B多CMS Agent协作系统。支持Shopify/WooCommerce/Magento三大平台，A2A架构协调库存/采购/财务/物流四大专家Agent，三层安全网保障审批合规。即装即用，零配置开箱。 Tags: latest:3.0.4 Version history: v3.0.4 | 2026-05-01T15:03:39.752Z | user Security fixes: removed eval/exec patterns, replaced exposed API key placeholders, cleaned prompt patterns v3.0.3 | 2026-04-29T03:50:57.891Z | user Remove hardcoded key patterns from test","editorialQuality":{"score":100,"threshold":65,"status":"ready","wordCount":974,"uniquenessScore":59,"reasons":[]}},"media":{"evidence":{"source":"no-media","verified":false,"confidence":"low","updatedAt":"2026-10-11T17:56:25.652Z","emptyReason":"No screenshots, media assets, or demo links are available."},"primaryImageUrl":null,"mediaAssetCount":0,"assets":[],"demoUrl":null},"ownerResources":{"evidence":{"source":"unclaimed","verified":false,"confidence":"low","updatedAt":"2026-10-11T17:56:25.652Z","emptyReason":"This page has not been claimed by the agent owner."},"hasCustomPage":false,"customPageUpdatedAt":null,"customLinks":[],"structuredLinks":{"docsUrl":null,"demoUrl":null,"supportUrl":null,"pricingUrl":null,"statusUrl":null},"customPage":null},"relatedAgents":{"evidence":{"source":"protocol-neighbors","verified":false,"confidence":"medium","updatedAt":"2026-10-11T20:56:57.468Z","emptyReason":null},"items":[{"id":"8ebccd8e-3863-4187-8355-c3f14e1f9edf","entityType":"agent","canonicalPath":"/agent/iofficeai-aionui","slug":"iofficeai-aionui","name":"AionUi","description":"Free, local, open-source 24/7 Cowork app and OpenClaw for Gemini CLI, Claude Code, Codex, OpenCode, Qwen Code, Goose CLI, Auggie, and more | 🌟 Star if you like it!","url":"https://github.com/iOfficeAI/AionUi","homepage":"https://www.aionui.com","source":"GITHUB_REPOS","protocols":["MCP","OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-10-09T19:11:12.944Z","createdAt":"2026-02-25T03:38:16.584Z","downloads":null},{"id":"b917f68a-ebff-438e-84f8-3f4b2494c0bc","entityType":"agent","canonicalPath":"/agent/activepieces-activepieces","slug":"activepieces-activepieces","name":"activepieces","description":"AI Agents & MCPs & AI Workflow Automation • (~400 MCP servers for AI agents) • AI Automation / AI Agent with MCPs • AI Workflows & AI Agents • MCPs for AI Agents","url":"https://github.com/activepieces/activepieces","homepage":"https://www.activepieces.com","source":"GITHUB_REPOS","protocols":["OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-04-15T02:22:12.426Z","createdAt":"2026-02-25T03:38:12.412Z","downloads":null},{"id":"5cb26759-3a39-483f-94cf-276a98c13bb8","entityType":"agent","canonicalPath":"/agent/cherryhq-cherry-studio","slug":"cherryhq-cherry-studio","name":"cherry-studio","description":"AI productivity studio with smart chat, autonomous agents, and 300+ assistants. Unified access to frontier LLMs","url":"https://github.com/CherryHQ/cherry-studio","homepage":"https://cherry-ai.com","source":"GITHUB_REPOS","protocols":["MCP","OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-04-11T14:38:40.986Z","createdAt":"2026-02-25T03:38:19.379Z","downloads":null},{"id":"6f6582d0-5d76-4f0f-b81d-86520247950b","entityType":"agent","canonicalPath":"/agent/copilotkit-copilotkit","slug":"copilotkit-copilotkit","name":"CopilotKit","description":"The Frontend for Agents & Generative UI. React + Angular","url":"https://github.com/CopilotKit/CopilotKit","homepage":"https://docs.copilotkit.ai","source":"GITHUB_REPOS","protocols":["OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-03-25T09:50:57.846Z","createdAt":"2026-02-25T03:39:14.617Z","downloads":null}],"links":{"hub":"/agent","source":"/agent/source/clawhub","protocols":[{"label":"OpenClaw","href":"/agent/protocol/openclew"}]}}}