Back to Details

data360beast

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Data360 Beast

Data360 Beast

Data360 Beast turns Salesforce Data 360 work into a proof-driven execution loop. Use it whenever the user is designing, building, validating, executing, or debugging Data 360.
Beast is not only a knowledge system. It is an execution system rooted in proven knowledge with proof and learning loops. The skill outcome should be execution-ready for the target org, not just advice:
text
Execute in any authorized, compatible Data 360 org after preflight confirms
org, data space, permissions, available tools, and mutation approval.
Data360 Beast 将Salesforce Data 360相关工作转化为以验证为驱动的执行循环。当用户进行Data 360的设计、构建、验证、执行或调试工作时，均可使用本工具。
Beast不仅是一个知识系统，更是一个基于成熟知识、具备验证与学习循环的执行系统。本技能的输出结果应可直接在目标组织中执行，而非仅提供建议：
text
Execute in any authorized, compatible Data 360 org after preflight confirms
org, data space, permissions, available tools, and mutation approval.

Companion MCP Setup (do this first)

配套MCP设置（请先完成）

Without these two servers the agent runs in degraded mode:
  1. sf-docs - fetches help.salesforce.com and developer.salesforce.com pages as clean Markdown. Install: docs/mcp-dependencies.md.
  2. data360 - exposes Data 360 Connect API tools (search, payload_examples, execute). Install: docs/mcp-dependencies.md.
若缺少以下两个服务器，代理将以降级模式运行：
  1. sf-docs - 抓取help.salesforce.com和developer.salesforce.com页面并转换为简洁的Markdown格式。安装指南：docs/mcp-dependencies.md。
  2. data360 - 提供Data 360 Connect API工具（搜索、payload_examples、执行）。安装指南：docs/mcp-dependencies.md。

Degraded-Mode Rules

降级模式规则

Apply these rules when one or both companion MCPs are unavailable.
When sf-docs is unavailable
  • Use only static references listed in this skill.
  • Do not state limits, quotas, feature availability, or billing figures from memory.
  • Label answers: source: docs-unverified (sf-docs unavailable)
  • Lower confidence to inferred for any claim requiring a live doc check.
When data360 MCP is unavailable
  • Do not attempt live Connect API calls or org validation.
  • Generate payloads from OpenAPI shape and proof ledger evidence only.
  • Label answers: validation: live-validation-unavailable (data360 MCP unavailable)
  • Do not fabricate returned IDs, status codes, or org-specific behavior.
When both are unavailable
  • Restrict to architecture guidance and structural payload design from static references.
  • Open every response with: [degraded mode: static references only]
  • Stop rather than guess for: field names not in the OpenAPI spec, current feature availability, limit values, connector behavior, and production mutation steps.
Stop condition
Stop and ask for authoritative input when: both MCPs are unavailable, the fact is absent from every static reference, and a wrong answer would cause data loss, an incorrect production mutation, or a security boundary violation.
当一个或两个配套MCP不可用时，需遵循以下规则：
当sf-docs不可用时
  • 仅使用本技能中列出的静态参考内容。
  • 不得凭记忆陈述限制、配额、功能可用性或计费数据。
  • 需为答案添加标签：source: docs-unverified (sf-docs unavailable)
  • 对于任何需要实时文档验证的声明，将置信度标记为“推断”。
当data360 MCP不可用时
  • 不得尝试发起实时Connect API调用或组织验证。
  • 仅基于OpenAPI结构和验证账本证据生成负载。
  • 需为答案添加标签：validation: live-validation-unavailable (data360 MCP unavailable)
  • 不得编造返回的ID、状态码或组织特定行为。
当两者均不可用时
  • 仅提供基于静态参考的架构指导和结构化负载设计。
  • 所有回复开头需标注：[degraded mode: static references only]
  • 对于以下内容，停止操作而非猜测：OpenAPI规范中未包含的字段名、当前功能可用性、限制值、连接器行为以及生产环境变更步骤。
停止条件
当以下情况同时发生时，停止操作并请求权威输入：两个MCP均不可用、所有静态参考中均无相关事实、且错误答案可能导致数据丢失、不正确的生产环境变更或安全边界违规。

Operating Loop

执行循环

Use the four-layer loop from docs/operating-model.md:
text
Think -> Act -> Prove -> Learn
  1. Think with Beast preflight, phase proof matrix, official docs, OpenAPI, Beast specialists, proof ledger evidence, and optional Salesforce 
    sf-skills
    companion guidance.
  2. Act only after authorization is clear, using the right MCP, direct 
    sf
    REST, 
    sf data360
    CLI, metadata/deployment tool, or helper script.
  3. Prove with returned ID, readback, status, count, data space, metadata, query result, log, event, or delivery signal.
  4. Learn by promoting only distilled public-safe lessons into Beast or routing scenario-heavy work to Labs.
For limits, use docs/data360/limits-source-precedence.md: current Data 360 limits first, Data Services usage when referenced, and legacy Customer Data Platform limits only when explicitly in scope.
End non-trivial responses as one of: execution-ready plan, executed-and-proven result, done-but-waiting result, blocked-with-reason, or lab-required.
使用docs/operating-model.md中的四层循环：
text
Think -> Act -> Prove -> Learn
  1. 思考：借助Beast预检、阶段验证矩阵、官方文档、OpenAPI、Beast专家、验证账本证据以及可选的Salesforce 
    sf-skills
    配套指导。
  2. 执行：仅在明确获得授权后执行，使用合适的MCP、直接
    sf
    REST接口、
    sf data360
    CLI工具、元数据/部署工具或辅助脚本。
  3. 验证：通过返回的ID、回读结果、状态、计数、数据空间、元数据、查询结果、日志、事件或交付信号进行验证。
  4. 学习：仅将提炼后的公开安全经验推广至Beast，或将场景密集型工作转至实验室处理。
关于限制，请参考docs/data360/limits-source-precedence.md：优先使用当前Data 360限制，引用时使用数据服务使用限制，仅在明确涉及相关范围时使用旧版客户数据平台（Customer Data Platform）限制。
非琐碎回复需以以下形式之一结尾：可执行计划、已执行并验证的结果、已完成但待确认的结果、受阻及原因说明、或需实验室支持。

Source Hierarchy

来源层级

  1. User-provided target org, files, data space, API version, and business goal.
  2. Official Salesforce docs fetched on demand.
  3. User-supplied or locally available Data 360 Connect API OpenAPI spec.
  4. Data360 Beast public docs and proof ledger.
  5. Live org validation in an explicitly authorized org.
Companion MCP install paths are documented in docs/mcp-dependencies.md. Do not assume 
sf-docs
or 
data360
MCP servers are bundled with the skill pack.
  1. 用户提供的目标组织、文件、数据空间、API版本及业务目标。
  2. 按需抓取的Salesforce官方文档。
  3. 用户提供或本地可用的Data 360 Connect API OpenAPI规范。
  4. Data360 Beast公开文档及验证账本。
  5. 在明确授权的组织中进行的实时组织验证。
配套MCP的安装路径记录在docs/mcp-dependencies.md中。请勿假设
sf-docs
或
data360
MCP服务器已随技能包捆绑提供。

Salesforce sf-skills Data 360 Companion

Salesforce sf-skills Data 360配套工具

After installing Beast, agents can optionally install the Salesforce 
forcedotcom/sf-skills
Data 360 companion subset for 
sf data360
execution playbooks, readiness checks, command templates, and CLI troubleshooting.
Use the companion only after Beast preflight and phase routing. Beast remains the authority for source order, proof labels, limits precedence, official-doc routing, OpenAPI discipline, confidence, and Labs promotion. 
sf-docs
remains the official-doc path for current Salesforce setup, permission, limit, licensing, and behavior claims.
The companion helps with command planning and runtime gotchas. Actual org actions still happen through approved CLI, MCP, API, metadata, or helper tools inside the preflight authorization boundary.
Install or dry-run the companion with:
bash
python3 skills/data360beast/scripts/install_sf_skills_data360_companion.py --dry-run
The companion contract is documented in docs/data360/sf-skills-data360-companion.md and the router reference is references/sf-skills-data360-companion.md. Use Data 360 in Beast-facing guidance; preserve exact Salesforce 
sf-skills
names only for folders, install paths, package metadata, or command surfaces.
Public-safe operating references:
  • scripts/data360beast.py: portable desired-state preflight, plan, apply, and knowledge-query runtime. Planning is safe by default; mutation requires an explicit execution flag and production approval.
  • docs/data360/deployment-runtime.md: implementation-spec, deployment-plan, run-state, certification, portability, and approval contract.
  • docs/data360/knowledge/claims.json: source-backed public claims, caveats, proof status, and owning specialist skill.
  • docs/beast-preflight.md: required envelope, tool inventory, proof target, and mutation gate.
  • docs/phase-proof-matrix.json: machine-readable phase routing, required sources, proof targets, and forbidden assumptions.
  • docs/phase-coverage-matrix.json: machine-readable coverage by phase, source, proof, helper, and frontier.
  • docs/data360/help/index.md: indexed Salesforce Help coverage.
  • docs/data360/developer/index.md: Developer Center coverage across Data 360 guide and reference families.
  • docs/data360/developer/learning-map.md: developer-guide synthesis for routing and skill updates.
  • docs/data360/model-gallery-implementation-map.md: public Data 360 model-gallery synthesis for DMO anchors, relationship paths, model grain, and implementation traps.
  • docs/data360/limits-source-precedence.md: current Data 360 limits first, Data Services usage follow-through, and legacy CDP comparison rules.
  • docs/data360/rag-search-index-retriever-playbook.md: public-safe RAG, search-index, chunking, retriever, and troubleshooting playbook distilled from a 45-page Salesforce public-facing best-practices PDF.
  • docs/data360/cost-usage-sizing-contract.md: cost and usage sizing contract for ingestion, query, insights, RAG, segmentation, activation, automation, and environment replay.
  • docs/data360/develop-package-deployment-matrix.md: develop, package, data kit, deploy, and readback matrix.
  • docs/data360/mcp-tool-selection.md: tool-selection guidance for 
    sf-docs
    , 
    data360
    , 
    datacloud-mcp-query
    , and direct 
    sf
    REST calls.
  • docs/data360/sf-skills-data360-companion.md: Salesforce 
    sf-skills
    Data 360 companion contract for optional 
    sf data360
    execution playbooks.
  • docs/data360/docs-watch-operating-model.md: weekly official-doc refresh, oversized Help fallback, audit, skill-sync, and GitHub publishing model.
  • docs/proof-ledger.md: public-safe evidence, caveats, confidence labels, labs references, and promotion status.
  • docs/labs-interface.md: boundary between Beast and the Labs proving ground for golden scenarios, synthetic journeys, payload experiments, traces, and future cookbook candidates.
  • docs/beast-evals.md: lightweight eval checks for future Beast mode changes.
Do not treat this repository as official Salesforce documentation. Do not hallucinate endpoint paths, payload fields, limits, permissions, or feature availability.
For implementation work, create an 
implementation-spec/v1
, run the runtime preflight and plan commands, and execute only after target capability discovery and mutation approval. Treat unsupported surfaces as blocked or manual handoff; never turn a plan into a production mutation by assumption.
安装Beast后，代理可选择安装Salesforce 
forcedotcom/sf-skills
的Data 360配套子集，用于
sf data360
执行剧本、就绪检查、命令模板及CLI故障排查。
仅在完成Beast预检和阶段路由后使用该配套工具。Beast在来源顺序、验证标签、限制优先级、官方文档路由、OpenAPI规范、置信度及实验室推广方面仍为权威。
sf-docs
仍是获取当前Salesforce设置、权限、限制、许可及行为声明的官方文档路径。
该配套工具可协助进行命令规划和运行时问题排查。实际组织操作仍需通过预检授权边界内的已批准CLI、MCP、API、元数据或辅助工具执行。
通过以下命令安装或试运行该配套工具：
bash
python3 skills/data360beast/scripts/install_sf_skills_data360_companion.py --dry-run
配套工具的协议记录在docs/data360/sf-skills-data360-companion.md中，路由参考记录在references/sf-skills-data360-companion.md中。在面向Beast的指导中使用Data 360；仅在文件夹、安装路径、包元数据或命令界面中保留Salesforce 
sf-skills
的准确名称。
公开安全的操作参考：
  • scripts/data360beast.py：可移植的期望状态预检、规划、应用及知识查询运行时。默认情况下规划操作是安全的；变更操作需明确的执行标志及生产环境批准。
  • docs/data360/deployment-runtime.md：实施规范、部署计划、运行状态、认证、可移植性及批准协议。
  • docs/data360/knowledge/claims.json：基于来源的公开声明、注意事项、验证状态及负责的专家技能。
  • docs/beast-preflight.md：必填的信封信息、工具清单、验证目标及变更闸门。
  • docs/phase-proof-matrix.json：机器可读的阶段路由、所需来源、验证目标及禁止假设。
  • docs/phase-coverage-matrix.json：机器可读的按阶段、来源、验证、辅助工具及前沿领域划分的覆盖范围。
  • docs/data360/help/index.md：索引化的Salesforce帮助文档覆盖范围。
  • docs/data360/developer/index.md：Data 360指南及参考系列的开发者中心覆盖范围。
  • docs/data360/developer/learning-map.md：用于路由和技能更新的开发者指南汇总。
  • docs/data360/model-gallery-implementation-map.md：公开Data 360模型库汇总，用于DMO锚点、关系路径、模型粒度及实施陷阱排查。
  • docs/data360/limits-source-precedence.md：优先使用当前Data 360限制，遵循数据服务使用规则，以及旧版CDP对比规则。
  • docs/data360/rag-search-index-retriever-playbook.md：基于45页Salesforce公开最佳实践PDF提炼的公开安全RAG、搜索索引、分块、检索器及故障排查剧本。
  • docs/data360/cost-usage-sizing-contract.md：用于 ingestion、查询、洞察、RAG、分群、激活、自动化及环境重放的成本和使用规模协议。
  • docs/data360/develop-package-deployment-matrix.md：开发、打包、数据套件、部署及回读矩阵。
  • docs/data360/mcp-tool-selection.md：
    sf-docs
    、
    data360
    、
    datacloud-mcp-query
    及直接
    sf
    REST调用的工具选择指南。
  • docs/data360/sf-skills-data360-companion.md：Salesforce 
    sf-skills
    Data 360配套工具协议，用于可选的
    sf data360
    执行剧本。
  • docs/data360/docs-watch-operating-model.md：每周官方文档更新、超大帮助文档回退、审计、技能同步及GitHub发布模型。
  • docs/proof-ledger.md：公开安全的证据、注意事项、置信度标签、实验室参考及推广状态。
  • docs/labs-interface.md：Beast与实验室验证环境之间的边界，用于黄金场景、合成流程、负载实验、追踪及未来候选手册。
  • docs/beast-evals.md：用于未来Beast模式变更的轻量级评估检查。
请勿将本仓库视为Salesforce官方文档。请勿编造端点路径、负载字段、限制、权限或功能可用性。
对于实施工作，请创建
implementation-spec/v1
，运行运行时预检和规划命令，并仅在完成目标能力发现和变更批准后执行。将不支持的表面视为受阻或手动交接；切勿仅凭假设将计划转化为生产环境变更。

Beast Preflight

Beast预检

For non-trivial work, capture or infer:
  • business goal and target phase
  • target org or alias, API version, data space, license posture, and user persona
  • asset lifecycle: disposable lab, sandbox, or production
  • authorization boundary: docs-only, metadata read, live validation, create or update, or production mutation
  • available thinking sources: 
    sf-docs
    , OpenAPI/Swagger, Beast specialists, proof ledger, optional Salesforce 
    sf-skills
    Data 360 companion, and user-provided files
  • available action/proof tools: 
    data360
    MCP, 
    datacloud-mcp-query
    , direct 
    sf
    REST, 
    sf data360
    CLI commands planned through the companion, metadata/deployment tools, and helper scripts
  • proof target and confidence label
If a required fact is missing, continue only when the task can safely proceed with an explicit assumption. Lower confidence instead of filling gaps with guesses.
对于非琐碎工作，需捕获或推断以下信息：
  • 业务目标及目标阶段
  • 目标组织或别名、API版本、数据空间、许可状态及用户角色
  • 资产生命周期：可丢弃实验室、沙箱或生产环境
  • 授权边界：仅文档、元数据读取、实时验证、创建或更新、或生产环境变更
  • 可用思考来源：
    sf-docs
    、OpenAPI/Swagger、Beast专家、验证账本、可选的Salesforce 
    sf-skills
    Data 360配套工具及用户提供的文件
  • 可用执行/验证工具：
    data360
    MCP、
    datacloud-mcp-query
    、直接
    sf
    REST接口、通过配套工具规划的
    sf data360
    CLI命令、元数据/部署工具及辅助脚本
  • 验证目标及置信度标签
若缺少必要事实，仅当任务可基于明确假设安全推进时才可继续。降低置信度，而非通过猜测填补空白。

Phase Router

阶段路由

Use docs/phase-proof-matrix.json as the machine-readable routing table. The bullets below are the human shorthand.
  • Connect: connectors, connections, connector schema, refresh, ingestion setup, connector permissions.
  • Prepare: data streams, DLOs, transforms, ingestion status, document processing.
  • Harmonize: DMOs, mappings, relationships, identity resolution, data graphs, model-gallery subject areas, anchor DMO selection, and relationship-path design.
  • Govern: data spaces, tags, classifications, security policies, access.
  • Retrieve: Query SQL, Query v2, Profile API, metadata discovery.
  • Insight: calculated insights, streaming insights, SQL validation.
  • Semantic: C360 semantic models, metrics, dimensions, shared dimensions.
  • AI/Search: search indexes, retrievers, unstructured retrieval, grounding.
  • Segment: segment create/read/publish/count, DBT segments, membership proof.
  • Act: activation targets, activations, data action targets, data actions.
  • Automation: triggered flows, DataObjectDataChgEvent, activation-triggered flows, refresh cadence, monitoring.
  • Develop/Package: API surface selection, External Client App auth, development environments, data kits, packageability, sandbox-to-production deployment, metadata coverage, cost and usage.
使用docs/phase-proof-matrix.json作为机器可读路由表。以下是简化的人工说明：
  • 连接（Connect）：连接器、连接、连接器 schema、刷新、 ingestion 设置、连接器权限。
  • 准备（Prepare）：数据流、DLO、转换、 ingestion 状态、文档处理。
  • 协调（Harmonize）：DMO、映射、关系、身份解析、数据图谱、模型库主题领域、锚点DMO选择及关系路径设计。
  • 治理（Govern）：数据空间、标签、分类、安全策略、访问权限。
  • 检索（Retrieve）：SQL查询、Query v2、Profile API、元数据发现。
  • 洞察（Insight）：计算洞察、流洞察、SQL验证。
  • 语义层（Semantic）：C360语义模型、指标、维度、共享维度。
  • AI/搜索（AI/Search）：搜索索引、检索器、非结构化检索、基础模型对接。
  • 分群（Segment）：分群的创建/读取/发布/计数、DBT分群、成员资格验证。
  • 执行（Act）：激活目标、激活操作、数据操作目标、数据操作。
  • 自动化（Automation）：触发流、DataObjectDataChgEvent、激活触发流、刷新节奏、监控。
  • 开发/打包（Develop/Package）：API表面选择、外部客户端应用认证、开发环境、数据套件、可打包性、沙箱到生产环境部署、元数据覆盖范围、成本及使用情况。

Engine-Aware Triage

引擎感知分流

Use docs/data360/architecture-engine-map.md when the work involves architecture, performance, cross-surface mismatches, or ambiguous troubleshooting.
Use docs/data360/interoperability-decision-map.md when the work involves external lakehouses, zero copy, ingestion strategy, freshness, cost/I/O, or hybrid architecture.
Use docs/data360/model-gallery-implementation-map.md when the work involves Data 360 data models, DMO choice, model diagrams, subject areas such as Case, Party, Privacy, Engagement, Product, Sales Order, Financial Services, Healthcare Provider, Vehicle Charger and Telematics, Student Financial Aid, or GenAI Audit and Feedback.
Use docs/data360/rag-search-index-retriever-playbook.md when the work involves RAG architecture, Agentforce Data Libraries, manual search indexes, chunking, hybrid/vector search, retriever filters, prompt grounding, Flow/Apex RAG orchestration, or retrieval troubleshooting.
Default loop:
text
symptom -> surface -> DSO/DLO/DMO/CIO/graph layer -> likely execution plane -> proof path
Interoperability loop:
text
workload -> freshness need -> governance need -> cost/I/O profile -> access pattern -> integration pattern -> proof path
Treat engine mapping as inferred unless proven in the target org. Spark-like processing, Trino-like query, Hyper-like analytics serving, Airflow-like orchestration, and Iceberg-style storage are useful mental models, not public API guarantees.
当工作涉及架构、性能、跨表面不匹配或模糊故障排查时，使用docs/data360/architecture-engine-map.md。
当工作涉及外部数据湖、零拷贝、 ingestion 策略、新鲜度、成本/I/O或混合架构时，使用docs/data360/interoperability-decision-map.md。
当工作涉及Data 360数据模型、DMO选择、模型图、主题领域（如案例、参与方、隐私、互动、产品、销售订单、金融服务、医疗保健提供商、车辆充电与远程信息处理、学生资助或生成式AI审计与反馈）时，使用docs/data360/model-gallery-implementation-map.md。
当工作涉及RAG架构、Agentforce数据库、手动搜索索引、分块、混合/向量搜索、检索器过滤器、提示词基础模型对接、Flow/Apex RAG编排或检索故障排查时，使用docs/data360/rag-search-index-retriever-playbook.md。
默认循环：
text
symptom -> surface -> DSO/DLO/DMO/CIO/graph layer -> likely execution plane -> proof path
互操作性循环：
text
workload -> freshness need -> governance need -> cost/I/O profile -> access pattern -> integration pattern -> proof path
除非在目标组织中得到验证，否则将引擎映射视为推断结果。类Spark处理、类Trino查询、类Hyper分析服务、类Airflow编排及类Iceberg存储是有用的思维模型，而非公开API保证。

Specialist Skill Routing

专家技能路由

After loading this router, use the matching specialist skill when the work is specific:
  • sf-datacloud: cross-phase architecture.
  • sf-datacloud-connectapi: REST, OpenAPI, Apex ConnectApi, payload design.
  • sf-datacloud-connect: connectors and connections.
  • sf-datacloud-prepare: streams, DLOs, transforms.
  • sf-datacloud-harmonize: DMOs, mappings, identity, data graphs.
  • sf-datacloud-governance: data spaces, access, tags, masking, policies.
  • sf-datacloud-retrieve: SQL, profile, metadata, query tools.
  • sf-datacloud-calculated-insights: calculated and streaming insight SQL.
  • sf-datacloud-segment: segments, counts, publish proof.
  • sf-datacloud-act: activations and data action delivery.
  • sf-datacloud-automation: flows, events, monitoring.
  • sf-datacloud-semantic-layer: semantic models and metrics.
  • sf-datacloud-ai-models: AI models and model outputs.
  • sf-datacloud-unstructured-retrieval: search indexes and retrievers.
  • sf-datacloud-analytics: reports, dashboards, analytics.
  • sf-datacloud-metadata-agentic: metadata semantics for agents.
加载本路由后，当工作具有特定性时，使用匹配的专家技能：
  • sf-datacloud：跨阶段架构。
  • sf-datacloud-connectapi：REST、OpenAPI、Apex ConnectApi、负载设计。
  • sf-datacloud-connect：连接器及连接。
  • sf-datacloud-prepare：流、DLO、转换。
  • sf-datacloud-harmonize：DMO、映射、身份、数据图谱。
  • sf-datacloud-governance：数据空间、访问权限、标签、掩码、策略。
  • sf-datacloud-retrieve：SQL、配置文件、元数据、查询工具。
  • sf-datacloud-calculated-insights：计算及流洞察SQL。
  • sf-datacloud-segment：分群、计数、发布验证。
  • sf-datacloud-act：激活及数据操作交付。
  • sf-datacloud-automation：流、事件、监控。
  • sf-datacloud-semantic-layer：语义模型及指标。
  • sf-datacloud-ai-models：AI模型及模型输出。
  • sf-datacloud-unstructured-retrieval：搜索索引及检索器。
  • sf-datacloud-analytics：报表、仪表盘、分析。
  • sf-datacloud-metadata-agentic：面向代理的元数据语义。

Connect API Workflow

Connect API工作流

For API work:
  1. Require or infer the target data space and API version.
  2. Search the OpenAPI spec for the operation family before writing payloads.
  3. Use the smallest valid payload first.
  4. Create only disposable lab assets unless the user explicitly asks for production changes.
  5. Read back by returned ID and check status/count/metadata.
  6. Capture caveats in the response so the next agent does not rediscover them.
对于API工作：
  1. 要求或推断目标数据空间及API版本。
  2. 在编写负载前，搜索OpenAPI规范以找到对应的操作族。
  3. 优先使用最小的有效负载。
  4. 仅创建可丢弃的实验室资产，除非用户明确要求进行生产环境变更。
  5. 通过返回的ID进行回读，并检查状态/计数/元数据。
  6. 在响应中记录注意事项，以便后续代理无需重新发现这些问题。

Proof Kernel

验证核心

Use the specialist skills and public docs for details. Keep this top router small and enforce these cross-cutting rules:
  • Use docs/phase-proof-matrix.json to pick the required source and proof target before answering.
  • Query success does not prove segment, transform, analytics, activation, or orchestration behavior; proof must match the target surface.
  • Use the model-gallery map before choosing DMO anchors, model grain, relationship paths, custom DMOs, or Data Graph shape.
  • Diagnose RAG as an offline chain and an online chain: source, chunk, index, retrieve, hydrate, prompt, generate, cite, and enforce access.
  • Choose external data architecture before payloads: ingestion, real-time ingestion, streaming ingestion, batch ingestion, live query, accelerated query, file federation, or hybrid.
  • Use the cost and usage sizing contract when a recommendation affects ingestion volume, query volume, refresh cadence, RAG indexing, AI processing, activation payloads, or multi-environment replay.
  • For limits, quotas, connector counts, licensing, billing, and feature availability, default to current Data 360 Limits and Guidelines. Follow Data Services Billable Usage Types when current docs route there. Treat Customer Data Platform limits as legacy unless the user explicitly says CDP/Customer Data Platform, the target org is known to use that license, or the answer is a labeled comparison.
  • Choose the API surface before coding: Connect REST, Apex 
    ConnectApi
    , Data 360 API / Direct API, SOQL-supported paths, Metadata API, or data kits.
  • Promote evidence only when it has official docs or OpenAPI shape, a minimal payload or command, readback proof, caveats, and failure modes.
  • Cost is a design gate: query only required fields, filter early, ingest selectively, aggregate before ingest when raw detail is unnecessary, and keep test data small.
  • Use 
    datacloud-mcp-query
    only as a retrieve-phase accelerator for Query SQL, list tables, and describe table. Use 
    data360
    MCP or direct REST for broader Connect API operations.
使用专家技能及公开文档获取详细信息。保持本顶层路由简洁，并执行以下跨领域规则：
  • 在回答前，使用docs/phase-proof-matrix.json选择所需来源及验证目标。
  • 查询成功并不证明分群、转换、分析、激活或编排行为；验证必须匹配目标表面。
  • 在选择DMO锚点、模型粒度、关系路径、自定义DMO或数据图谱形状前，使用模型库映射。
  • 将RAG诊断为离线链和在线链：来源、分块、索引、检索、填充、提示词、生成、引用及访问控制。
  • 在编写负载前选择外部数据架构： ingestion、实时 ingestion、流 ingestion、批量 ingestion、实时查询、加速查询、文件联邦或混合架构。
  • 当建议影响 ingestion 量、查询量、刷新节奏、RAG索引、AI处理、激活负载或多环境重放时，使用成本和使用规模协议。
  • 对于限制、配额、连接器数量、许可、计费及功能可用性，默认使用当前Data 360限制及指南。当当前文档指向数据服务计费使用类型时，遵循该类型。除非用户明确提及CDP/Customer Data Platform、目标组织已知使用该许可，或答案为带标签的对比内容，否则将客户数据平台限制视为旧版内容。
  • 在编码前选择API表面：Connect REST、Apex 
    ConnectApi
    、Data 360 API / Direct API、SOQL支持路径、Metadata API或数据套件。
  • 仅当证据具备官方文档或OpenAPI结构、最小负载或命令、回读验证、注意事项及故障模式时，才可推广该证据。
  • 成本是设计闸门：仅查询所需字段、尽早过滤、选择性 ingestion、当原始细节不必要时在 ingestion 前进行聚合、并保持测试数据量较小。
  • 仅将
    datacloud-mcp-query
    用作检索阶段的加速器，用于SQL查询、列表表格及描述表格。对于更广泛的Connect API操作，使用
    data360
    MCP或直接REST接口。

Output Contract

输出协议

Prefer answers with this shape:
text
Preflight
Recommendation
Source path: documented | tested | inferred
Payload/command/query
Validation readback
Caveats
Next proof step
If the task is customer-facing, keep the answer sharp and avoid internal lab details unless they directly support the recommendation.
优先采用以下结构的答案：
text
预检
建议
来源路径：已文档化 | 已测试 | 推断
负载/命令/查询
验证回读
注意事项
下一步验证步骤
若任务面向客户，需保持答案简洁，避免提及内部实验室细节，除非这些细节直接支持建议。