Marketing OS

AI Agent 营销操作系统 — 包含 Virtual CMO（战略大脑）和 Marketing Operator（执行引擎），提供市场分析、策略制定、Campaign 规划与执行追踪全链路能力。

Security Scan

The package is an instruction-only, self-contained Marketing OS whose declared purpose matches the files and runtime instructions; it does not request extra credentials or install code, but you should review adapter endpoints, credential usage, and memory retention before enabling external integrations or auto-mode.

Purpose & Capability

Name/description (Virtual CMO + Marketing Operator) align with the provided prompts, schemas, workflows, and memory files. Nothing in the bundle asks for unrelated capabilities (no AWS/DB credentials, no unexpected binaries). Adapters are described but disabled by default in configs; they are appropriate for the described integration points.

Instruction Scope

SKILL.md and included prompts instruct the agent to read/write local memory (memory/*.json) and logs, to load prompts, validate I/O against schemas, and call adapters when configured. Reading/writing memory and logs is coherent with a campaign-management skill; however, this is a data-handling surface the user should review (what is persisted, retention settings, and PII handling). Adapters mention external APIs and credential references — those will only be used if the user configures endpoints and credentials.

Install Mechanism

Instruction-only skill with no install spec and no code files to execute. No remote downloads or package installs are present, which minimizes supply-chain risk.

Credentials

The skill declares no required environment variables, no primary credential, and no config paths outside its own package. Adapter docs reference auth methods (api_key, oauth2) and credentials_ref placeholders, which are expected for optional external integrations but are not demanded by the skill as shipped.

Persistence & Privilege

always:false and default autonomous invocation are unchanged (normal). The skill explicitly persists to memory/*.json and logs/*.json; this is legitimate for its purpose but increases data persistence surface. The runtime config includes gates (requires_human_approval entries) to prevent dangerous automatic actions like external_api_integration or launching paid campaigns when correctly applied.

Assessment

This package appears to do what it says: it is an instruction-only Marketing OS that reads/writes local memory and logs, uses structured prompts and schemas, and defines optional adapters for external APIs. Before enabling it in production: 1) Keep auto_mode disabled initially and test workflows manually. 2) Inspect and edit configs/system.config.json — verify allowed_actions, requires_human_approval, memory retention, and adapters are set to disabled unless you intentionally configure them. 3) If you enable adapters, provide endpoints and credentials only to trusted services and store credentials in your secure store (the adapters reference credentials_ref). 4) Confirm PII handling: the CRM adapter warns PII should not be stored in memory; ensure your runtime enforces that. 5) Review logging and memory retention policies (memory/*.json and logs/*.json) to meet privacy/compliance needs. 6) Validate generated content and strategy outputs with human reviewers before publishing or spending budget. These precautions will reduce risk while using the skill.

🚀 Marketing OS — AI Agent 营销操作系统

概述

Marketing OS 是一个模块化、Schema 驱动的营销智能系统，设计为可直接接入任何 AI Agent Runtime 的 Skill Package。

它提供两个核心角色：

角色	功能
Virtual CMO	战略大脑 — 市场分析、机会识别、策略制定
Marketing Operator	执行引擎 — 任务分解、Campaign 管理、指标追踪

两个角色通过结构化协作协议 (schemas/cmo_to_operator.schema.json) 通信，确保战略到执行零歧义。

---

使用场景

场景	触发方式
市场发现	"分析当前市场机会并生成战略建议"
Offer 选择	"从已识别的机会中选择最佳产品匹配"
Campaign 规划	"把策略拆解为可执行任务"
执行冲刺	"执行所有任务并收集指标反馈"

---

技能调用格式

skill: marketing-os
input:
  mode: market_discovery | offer_selection | campaign_planning | execution_sprint
  business_context:
    company_name: "公司名称"
    products: ["产品1", "产品2"]
    target_market: "目标市场"
    budget_range: "预算范围"
    brand_positioning: "品牌定位"
  market_data:            # 可选 — 外部市场信号
    search_trends: []
    competitor_moves: []
    audience_signals: []
  mission_id: "xxx"       # campaign_planning / execution_sprint 时必填
  auto_mode: false        # 是否全自动执行

---

系统架构

┌─────────────────────────────────────────────────┐
│                  Agent Runtime                   │
├─────────────────────────────────────────────────┤
│                                                  │
│   ┌──────────────┐     ┌─────────────────────┐  │
│   │ Virtual CMO   │────▶│ Marketing Operator  │  │
│   │ (Strategy)    │◀────│ (Execution)         │  │
│   └──────┬───────┘     └──────────┬──────────┘  │
│          │                        │              │
│   ┌──────▼────────────────────────▼──────────┐  │
│   │              Shared Memory                │  │
│   │  (insights / campaigns / learnings)       │  │
│   └──────────────────────────────────────────┘  │
│          │                        │              │
│   ┌──────▼──────┐   ┌────────────▼───────────┐  │
│   │  Workflows   │   │      Adapters          │  │
│   │  (flow.json) │   │  (CRM/Data/Content)    │  │
│   └─────────────┘   └───────────────────────┘  │
└─────────────────────────────────────────────────┘

---

核心模块

1. 🧠 Virtual CMO — 战略分析

执行步骤：

Step 1: 收集市场数据（信号、趋势、竞争对手动态）
Step 2: 分析信号 — 分类、评分（signal_strength 1-10）
Step 3: 识别机会 — 聚类相关信号，评估市场规模/竞争/匹配度
Step 4: 计算优先级 — priority_score = (信号强度×0.3) + (市场规模×0.25) + (能力匹配×0.25) + (紧迫度×0.2)
Step 5: 生成策略 — 定位、渠道推荐、KPI、风险评估
Step 6: 输出任务简报 — 传递给 Marketing Operator

输出格式（schemas/cmo_output.schema.json）：

{
  "analysis_id": "UUID",
  "market_opportunities": [{"title": "...", "priority_score": 85, "confidence": "high"}],
  "target_segments": [{"name": "...", "pain_points": ["..."]}],
  "recommended_actions": [{"action": "...", "priority": "high", "owner": "operator"}],
  "risks": [{"description": "...", "severity": 7, "mitigation": "..."}],
  "next_steps": [{"action": "...", "owner": "operator", "deadline_type": "immediate"}],
  "confidence_level": 78
}

2. ⚙️ Marketing Operator — 任务执行

执行步骤：

Step 1: 验证 CMO Mission Brief
Step 2: 任务分解 — 将 action 拆成原子任务（每个任务有 owner/deadline/expected_result）
Step 3: 资源分配 — 预算、渠道、工具映射
Step 4: Campaign 组装 — 聚合任务，设置 KPI 目标
Step 5: 执行追踪 — 状态管理（pending → in_progress → completed/blocked/failed）
Step 6: 指标收集 — 量化（曝光/点击/转化）+ 定性（互动质量/品牌感知）
Step 7: 生成反馈 — 向 CMO 报告结果、学习、建议调整

---

协作协议

CMO → Operator 通信格式（schemas/cmo_to_operator.schema.json）：

{
  "mission_id": "UUID",
  "objective": "明确的可衡量目标",
  "target_audience": {"segment_name": "...", "pain_points": ["..."]},
  "strategy": {"positioning": "...", "approach": "..."},
  "priority": "critical | high | medium | low",
  "recommended_channels": [{"channel": "LinkedIn", "priority_rank": 1}],
  "actions": [{"action": "...", "priority": "high"}],
  "success_criteria": {"primary_kpi": {"metric": "conversions", "target": "100"}}
}

Operator → CMO 反馈格式（schemas/feedback.schema.json）：

{
  "feedback_id": "UUID",
  "execution_result": "事实总结",
  "metrics": {"impressions": 15000, "clicks": 450, "conversions": 23},
  "learnings": ["[MEASURED] LinkedIn 数据驱动标题 3x 互动率"],
  "recommendations": ["将 30% 预算从 Display 转移到 LinkedIn"]
}

---

行为规则

[!IMPORTANT] 系统内建以下严格约束：

• ❌ 不允许模糊建议（"考虑"、"或许"、"可以试试" 一律禁止）
• ✅ 必须给出优先级（critical / high / medium / low）
• ✅ 必须给出下一步行动
• ✅ 必须给出风险评估
• ✅ 必须区分 [FACT] / [INFERENCE] / [RECOMMENDATION]
• ✅ 信号强度 < 4 必须标记 "uncertain"
• ✅ 信息不足必须明确声明 "INSUFFICIENT DATA"

---

文件结构

marketing-os/
├── SKILL.md                          # 本文件
├── README.md                         # 系统文档
├── skills/
│   ├── virtual-cmo/                  # 战略分析
│   └── marketing-operator/           # 执行引擎
├── prompts/                          # 4 个结构化 LLM Prompt
├── schemas/                          # 5 个 JSON Schema
├── workflows/                        # 3 个工作流编排
├── memory/                           # 3 个持久化存储
├── logs/                             # 执行审计日志
├── configs/                          # 运行时配置
└── adapters/                         # 外部接口规范（CRM/Data/Content）

---

扩展方式

• 新增 Skill：在 skills/ 下创建目录，包含 skill.json + logic.md + system_prompt.txt
• 新增 Workflow：在 workflows/ 下创建 .flow.json
• 新增 Adapter：在 adapters/ 下创建 .adapter.md
• 接入 Stripe / CRM / 内容系统：配置 configs/system.config.json 中的 adapters 部分

[!TIP] 建议先以 auto_mode: false 手动运行各 workflow，验证输出质量后再开启自动模式。

---

Skill Version: 1.0.0 | Designed for AI Agent Runtime Integration | 2026-03-22