投资工作流

［何时使用］当用户需要投研分析时；当用户说'分析这个标的'、'现在什么值得买'、'对 XX 行业怎么看'、'这个热点有什么影响'、'开会讨论投资'时触发。场景驱动的投研全流程，覆盖 6 个场景。

Security Scan

Skill description and behavior mostly align with an investment-research workflow, but the runtime instructions depend on local tooling (mcp-aktools / data_layer) while the skill metadata declares no required binaries or install steps — that mismatch and the broad allowed-tool permissions merit caution.

Purpose & Capability

Name/description match the content: a scenario-based investment research workflow. However, SKILL.md and README refer to local tools (mcp-aktools at ~/.local/bin/mcp-aktools and a data_layer/ directory) and expect those providers; the registry metadata lists no required binaries, env vars, or install steps. This is an incoherence: the skill will realistically need those local tools to function but does not declare them.

Instruction Scope

SKILL.md describes calling shared skill modules and obtaining market/data via the local data layer or mcp-aktools; it does not instruct indiscriminate file access or secret harvesting. It does expect reading/writing report templates and invoking local binaries (exec/bash). Those actions are coherent with producing Markdown reports and querying a local data provider, provided the declared providers exist.

Install Mechanism

There is no install spec (instruction-only), which is low risk. But because the instructions explicitly rely on external binaries and a data_layer directory (not installed by the skill), the skill will fail or may try to exec arbitrary commands as part of fallbacks — the absence of a clear install or dependency declaration is a concern and reduces transparency.

Credentials

The skill requests no environment variables or credentials and references AKShare (a free data provider) via mcp-aktools/data_layer. That is proportionate to its stated purpose. Note: the documentation claims 'zero API Key' for mcp-aktools, so no secret exfiltration is requested by the skill itself.

Persistence & Privilege

Flags: always=false and user-invocable=true (normal). The SKILL.md permits use of Exec, Bash, Read, Write and WebSearch — justified for calling local tools and producing reports, but these permissions also allow arbitrary shell commands and filesystem writes when the agent executes the skill. The skill does not request to modify other skills or system-wide settings.

What to consider before installing

This skill appears to implement the stated investment-research workflow, but it relies on local tooling (mcp-aktools and a data_layer directory) that are referenced in docs but not declared in the registry metadata. Before installing or enabling: 1) ask the publisher/source for a homepage or source repository and a clear dependency/install list; 2) ensure mcp-aktools and data_layer are installed from trusted sources and inspect those tools (they run locally and may execute network calls); 3) test the skill in a sandbox or limited account first, because the skill is permitted to run shell commands and read/write files; 4) if you cannot verify the external tools, treat this skill as untrusted — the metadata omission (no required binaries declared) is the main red flag that justifies caution.

investment-workflow: 投资工作流 🎯

📋 功能描述

帮助用户系统化执行投研全流程。不是一套固定流程，是 6 个用户场景 × 共享 Skill 模块的排列组合。

适用场景：

• 买股票 / 投行业 / 扫描推荐 / 行业看法 / 热点分析 / 开会讨论

边界条件：

• 不替代深度基本面研究
• 输出为 Markdown 报告，需配合 data_layer / mcp-aktools 获取真实数据
• 场景识别依赖用户输入关键词
• 模糊输入处理：用户输入过于模糊（如"最近怎么样？"）时，先询问澄清（行业/事件/标的），不强行触发完整工作流

---

🔄 6 个核心场景

场景	触发词	调用步骤	输出
买股票	"我要买股票"、"分析 XX"	data-query → stock-research → decision-integrate	Markdown 决策报告
投行业	"我要投行业"、"XX 行业值得投吗"	industry-rank → data-query → multi-view → decision-integrate	Markdown 行业报告
扫描推荐	"现在什么值得买？"	market-scan → industry-rank → data-query → decision-integrate	Top 3 标的 + 优先级
行业看法	"对 XX 行业怎么看"	industry-rank → deep-think → plain-explain	Markdown 散文
热点分析	"这个热点有什么影响"	market-scan → industry-rank → multi-view → plain-explain	Markdown 影响分析
开会讨论	"开会讨论"	multi-view → decision-integrate	Markdown 会议纪要

详细共享 Skill 说明 → references/shared-skills.md 数据层集成说明 → references/data-layer-integration.md

---

⚠️ 常见错误

错误 1：线性执行所有阶段

问题：
• 用户只问"XX 怎么看"，却跑完 5 个阶段
• 输出冗长，用户找不到重点

解决：
✓ 严格按场景定义调用 2-4 个共享 Skill
✓ 输出聚焦场景核心问题

错误 2：忽略数据层降级

问题：
• mcp-aktools 或 data_layer 调用失败
• 报告数据为空或报错

解决：
✓ 优先调用 data_layer（带缓存）
✓ 失败时降级到本地缓存或明确告知"数据获取失败"
✓ 标注数据来源与时间

错误 3：推理过程缺失

问题：
• 直接给结论，用户不知道"怎么想到的"
• 缺乏降秩/追本/验证过程

解决：
✓ 每个结论必须保留推演链
✓ 使用 [数据：指标 | 来源：数据源 | 时间：时间] 格式
✓ 附 ASCII 关系图辅助理解

---

🧪 使用示例

输入：

分析消费 ETF 是否值得投？

预期输出：

• 识别场景：投行业
• 调用：industry-rank → data-query → multi-view → decision-integrate
• 输出：Markdown 报告（含降秩分析、数据验证、圆桌讨论、决策建议）

输入：

现在什么值得买？

预期输出：

• 识别场景：扫描推荐
• 调用：market-scan → industry-rank → data-query → decision-integrate
• 输出：Top 3 标的 + 优先级 + 逻辑

---

🔧 故障排查

问题	检查项
不触发	description 是否包含触发词？用户输入是否匹配场景？
数据为空	data_layer 是否安装？mcp-aktools 是否运行？缓存是否过期？
输出过长	是否跨场景调用？检查场景定义，只调用必要步骤
推理缺失	是否跳过降秩/追本步骤？检查共享 Skill 执行顺序

---

🔗 相关资源

• 共享 Skill 文档：references/shared-skills.md
• 数据层集成：references/data-layer-integration.md
• 报告模板：templates/report-template.md
• 标准参考：docs/SKILL-STANDARD-v3.md