muxuuu/serenity-skill

Serenity.skill – AI‑assisted investment‑research skill

What it is

A skill (plugin) for AI agents such as OpenAI Codex or Anthropic Claude Code. It encodes the “Serenity” research methodology – a structured way to turn market hot‑topics (AI chips, robotics, etc.) into a shortlist of stocks or funds that are truly close to a supply‑chain bottleneck.
The repository does not contain a model or training code. It ships a SKILL.md file, prompt templates, reference documents and a tiny validation script that let an agent load the skill and call it with natural‑language queries.

How it works

Input – the user asks the agent a question like “研究现在 A 股 AI 半导体产业链”。
Prompt generation – the skill inserts a pre‑written prompt (see assets/research-prompt-pack.md) that tells the model to:
- break the hot topic into downstream demand → system integration → chips/devices → equipment → materials → packaging → infrastructure;
- look for nodes with few suppliers, long verification cycles, capacity‑expansion difficulty, strict customer certification, high material purity;
- pull public data (announcements, filings, earnings, patents, tender documents, etc.) to verify the presence of those bottlenecks.
Reasoning – the LLM (running in the host client) performs the research, ranks candidate companies/funds, and returns a concise report that lists:
- the most promising 3‑5 targets,
- why each target is close to a “real” bottleneck,
- any missing evidence and the main risks.
Output – a plain‑text summary formatted exactly like the examples in the README (company name, bottleneck, evidence, risk, next‑step checklist).

What you can ask it to do

Situation	Example query you can give the agent	What Serenity.skill returns
You see a hype topic but don’t know where to start	`最近 AI 半导体很火，普通人应该先研究哪些方向？`	A hierarchy of the AI‑semiconductor supply chain and the top few segments worth researching.
You want to know which part of the robot supply chain is most constrained	`机器人产业链里，哪些环节更可能先出机会？`	Comparison of robot‑system, actuator, sensor, and gearbox layers with a ranking based on scarcity and expansion difficulty.
You suspect a stock is just “riding the hype”	`帮我挑战天孚通信是 CPO 核心供应商的说法`	A step‑by‑step fact‑check: customer list, order evidence, capacity announcements, and a risk assessment.
You need a short list of funds/ETFs that truly benefit from a theme	`机器人主题基金应该重点看哪些上游环节？`	The upstream segments that matter most and the funds whose top holdings align with those segments.
You have several candidate stocks and want a research priority list	`比较 A、B、C 三家公司，谁的上涨逻辑更清楚？`	A ranked table with bottleneck proximity, evidence strength, valuation pressure, and risk flags.

Installation & usage

The skill is just a directory of markdown files. To make it visible to an agent client, copy the whole folder into the client’s skill directory (e.g., ~/.agents/skills/serenity-skill for Codex or ~/.claude/skills/serenity-skill for Claude Code).
No Python runtime is required for the skill itself; a small helper script (scripts/validate_skill.py) can be run locally to verify the folder structure.

After installation, you invoke it with a slash command in the client, e.g.:

$serenity-skill 研究现在 A 股 AI 半导体产业链，说明优先研究方向、证据和反方理由。

/serenity-skill 挑战天孚通信是 CPO 核心供应商的说法，逐项核对原始披露。

What it does not do

It does not fetch real‑time market data, place trades, or guarantee any investment returns.
It does not contain any machine‑learning code; the heavy lifting is done by the underlying LLM that the host client provides.
The examples in the repo are snapshots up to 2026‑09‑14 and are not continuously updated.

Why it might be useful

Provides a repeatable, evidence‑first workflow for investors who want to leverage LLMs without relying on vague “buzz‑word” analysis.
Encapsulates a domain‑specific research methodology (the “Serenity” approach) into a reusable skill that can be called from any compatible agent platform.
Keeps the research process transparent: every recommendation is accompanied by the data sources and the logical steps that led to it.

License