Omni E-commerce Operations Expert Team

安装指南

1. ClawHub CLI 从源直接安装技能的最快方式。
   npx clawhub@latest install omni-ecom-skill
2. 手动安装 将技能文件夹复制到以下位置之一
   全局模式
   ~/.openclaw/skills/
   工作区
   /skills/
   优先级：工作区 > 本地 > 内置
3. 提示词安装 将此提示词复制到 OpenClaw 即可自动安装。
   请帮我使用 Clawhub 安装 omni-ecom-skill。如果尚未安装 Clawhub，请先安装（npm i -g clawhub）。
   更倾向于下载？
   获取包含原生技能文件的ZIP压缩包。
   下载技能 (.zip)

什么是

Omni-ecom is an e-commerce operations analysis skill for store diagnostics, weekly, monthly, and annual reports, promotion reviews, and advertising-spend profitability analysis. It works from user-provided original exports, without requiring platform APIs, and uses source recalculation, coverage checks, and independent review to make findings traceable.
For supported hosts, a real four-specialist workflow contributes to comprehensive reports, with each expert completing their own analysis before the team lead reconciles results. The skill distinguishes ROAS from profit ROI, avoids inventing missing costs or conclusions, and only presents final deliverables after required evidence, PDF, and completion checks pass.
Omni E-commerce Operations Expert Team 应用场景
Diagnose store performance using authorized platform exports.
Prepare weekly, monthly, or annual business reviews.
Review major promotions and identify evidence-backed findings.
Analyze advertising spend while distinguishing ROAS from profit ROI.
Reconcile metrics, coverage, and conflicting expert findings before report delivery.
Check whether supplied business files are sufficient to begin an analysis; if not, return an intake-blocked outcome rather than fabricate a report.
Omni E-commerce Operations Expert Team 工作原理
The user specifies the authorized client scope, platform, reporting period, task objective, and original exported files.
The skill checks host support and intake completeness, then identifies and classifies supplied materials. Empty, unknown, or insufficient inputs are accounted for rather than silently discarded.
Original files are read within the run scope and bound to the run, client, version, and source fingerprints. The source exports remain the basis for business facts.
For a comprehensive report, four specialists perform and return their own analyses using the supported host's task mechanism. Single-point handling is reserved for clearly scoped individual questions.
The team lead checks calculations, source coverage, analytical depth, and conflicts; claims and causal conclusions are checked against evidence. Missing costs or refund matches prevent unsupported net-profit or budget conclusions.
The report is assembled with task-appropriate conclusions, analysis, actions, and a traceable evidence appendix, then rendered as a chart-based PDF by default.
A separate reviewer checks figures, coverage, decisions, and every page of the frozen PDF. Delivery requires the required scope, claim, PDF, review, and completion gates to pass.
Omni E-commerce Operations Expert Team 配置指南
Obtain the ClawHub distribution package and read references/clawhub-distribution.md before running it. This package is not a WorkBuddy installation plugin and does not add subprocess tools to other hosts.
Confirm Python is available. The skill metadata lists python as a required binary.
Use a supported host adapter: the documented runtime currently implements WorkBuddy and Codex. Other hosts return host_adapter_not_available; do not treat them as supported multi-agent execution.
Read references/runtime-v2.md and references/workflow-refinement.md before handling a task. Follow the expert handoff contract and task cards referenced by the skill.
Provide the authorized client scope, platform, period, task purpose, and original business exports. Platform API credentials are not required.
Optionally configure a local client registry path or WorkBuddy recovery-check configuration directory if relevant. These are local configuration paths, not platform credentials.
python --version
OMNI_ECOM_CLIENT_REGISTRY is an optional path to a locally authorized client-name checklist; no client list is bundled or uploaded. CODEBUDDY_CONFIG_DIR is optional and used only for WorkBuddy recovery checks and development acceptance. No installation command or default configuration path is specified in the supplied skill document.
Omni E-commerce Operations Expert Team 数据架构与分类体系
The skill describes a run-scoped evidence and delivery process rather than a fixed business-data schema.
Data or artifact Role and handling
Original business exports Source of operational facts; read within the current run scope. Missing or unknown inputs are surfaced, not silently ignored.
Source copies and fingerprints Bind evidence to the current client, run, version, and source identity.
run-journal.sqlite and host_runtime Used for current run state; the skill identifies these as the current state sources.
Expert handoffs Each specialist supplies their own analysis; the program assembles fixed fields rather than inventing expert analysis.
Report and evidence appendix The default deliverable is a chart-based PDF with conclusions, analysis, actions, coverage, sources, and traceable expert contributions.
Review and completion evidence Records the independent review and required release/completion checks.
Reports identify the expert-team version, maintenance batch, and report revision. The stated formal version is v2.0.3, maintenance batch 20261014-r2; report revisions are tracked separately, such as R1 or R2. Historical materials retain their original identity and are read-only for migration; they do not inherit current approval, execution, or completion status.
Omni E-commerce Operations Expert Team 高级特性
Host-aware multi-agent execution:
WorkBuddy uses
TeamCreate
/
Agent
, while Codex uses
spawn_agent
/
wait_agent
; their tools and task credentials are not interchangeable. Other hosts are explicitly unsupported by the current runtime.
Four-specialist comprehensive analysis:
Comprehensive reports require real specialist contributions and completed handoffs before synthesis; the model choice does not change this requirement.
Evidence-first calculations:
Recalculation checks sources and formulas. Unknown costs are not treated as zero, and unsupported spend or ROI claims are avoided. ROAS and profit ROI are kept distinct.
Coverage and input safeguards:
Full identification, classification, dynamic report sections, and coverage checks account for empty tables, unknown file types, and insufficient data. No business files means an intake-blocked result, not a fabricated report.
Independent PDF release review:
A separate reviewer checks the frozen report's figures, coverage, decisions, and every PDF page. Matching file fingerprints alone cannot substitute for human review.
Explicit gates and recovery:
Scope checks, claim guard, PDF validation, independent release, and completion gates must pass before final delivery. Interrupted work resumes with its original task ID; retries are limited to roles with recorded failures or rejections.
Versioned delivery controls:
Formal and candidate versions are distinguished, and candidate acceptance does not certify a released customer deliverable. Only a released package with
release_ready=true
can be signed off as formally complete.

name: omni-ecom description: 全域电商经营专家团 v2.0.3；用于店铺诊断、周报/月报/年报、大促复盘和投流利润分析。识别完整授权报表，以真实多专家、数字来源复算与独立复核交付图表 PDF。 metadata: openclaw: requires: bins: [python] envVars: - name: OMNI_ECOM_CLIENT_REGISTRY required: false description: 用户明确授权的本机客户名称检查清单路径；不上传或内置客户清单。 - name: CODEBUDDY_CONFIG_DIR required: false description: 仅 WorkBuddy 恢复检查及开发验收时使用的本机配置目录，不是平台凭证。 全域电商经营专家团

ClawHub 分发版，沿用原条目 omni-ecom-skill，专家团名称仍为 omni-ecom。启动前先读 安装与宿主能力边界。本包不是 WorkBuddy 安装插件，也不自动为其他宿主增加子任务工具。当前运行协议仅实现 WorkBuddy 和 Codex；其他宿主先返回 host_adapter_not_available，不得扮演多个专家、伪造 ID 或宣布正式报告完成。

维护批次：20261014-r2；专家团仍为 v2.0.3。报告同时标注版本、维护批次和报告修订号；不将历史验收视为本批次新任务验收。

当前正式版：v2.0.3，上一正式版 v2.0.2。实际身份只取 version-info.json；候选版不得安装成正式版，也不能当客户交付。报告标注专家团版本及独立修订号 R1/R2，历史报告保持原身份。

用户提供客户范围、平台、期间、任务目的和原始导出报表即可；不需要平台 API。模型沿用用户选择。平台 API、后台动作/回读和真实跨模型路线基线均为 not_enabled；不把保留实验模块说成已启用。

主理人先读 宿主运行协议，按 流程修整 执行。岗位按自己的任务卡和 共同交接规范 处理，不加载无关可选说明。

必须保留：

全量识别、归类、动态章节及覆盖检查。空表、未知类型、资料不足均有去向和影响，不静默丢弃。没有业务文件只交 intake_blocked，不伪造报告或专家。
来源副本绑定本次客户、运行、版本与指纹。原始业务文件按 run-scope 限制读取；当前状态仅用 run-journal.sqlite 与 host_runtime。
真实原表是经营事实来源；未知费用不填零、不编造花费或 ROI。买家转化率为同口径支付买家/去重访客，GMV/访客叫访客价值；ROAS 与利润 ROI 分开，缺匹配退款及完整成本不出净利润和预算结论。数字、公式与归因按 来源复算 核对。
综合报告真实调用四位分析专家，原生完成和回传齐备后才并稿；WorkBuddy 用 TeamCreate/Agent，Codex 用 spawn_agent/wait_agent，工具和任务凭证不混用。模型选择不改变这些要求。明确单项问题才可走 single_point。
每位专家亲自填写分析，程序仅装配固定字段。首次 a1 符合正式 handoff；六题逐题有答案或有证据的限制，L2 拆解、必要的 L3/L4 和七项自检保留。每次只向团长回传一次，不与同级聊天。
团长核对数字、覆盖、深度与冲突；必要时只定向核查涉事岗位。因果判断按 因果证据，冲突裁决按 裁决交付，中断恢复用原任务 ID。只有已记录真实失败/拒收的岗位才可有限重试。
默认图表化 PDF。依据任务类型呈现结论、完整分析与行动；全量覆盖、来源和真实岗位贡献在证据附录可追踪。资料不足不补造经营图表或确定性结论。
先完成报告再冻结确切文件，由真实独立复核员核数字、覆盖、裁决和最终 PDF 每页；passed 且必改项为空才放行。程序指纹一致不能代替复核决定；复核员不改冻结文件。
公域范围检查、claim_guard、PDF验真、独立放行和 completion_gate 全部通过，且原生阶段已完成才展示最终交付。子任务未回传时主任务不能先完成。
新交接件、基线、数字索引、报告、复核、完成与阶段凭证均使用当前 2.0 契约并绑定生成版本；旧资料只读，不继承审批、执行或完成资格。旧资料迁移 仅在导入历史资料时读取。

发布 ZIP 包含 SKILL.md，通过 release_guard；真实发布按 双宿主验收 取得实际多专家、图表 PDF 和独立复核证据。候选验收使用 completion_gate.py --candidate-acceptance，状态 candidate_pipeline_complete；仅 released 且 release_ready=true 可签发正式完成，安装通过不代表客户报告完成。

WhatSync

sycm-shop-report