convez.ai
SkillFinanceiroAvançado
Mrjie7205

Serenity Bottleneck Hunter

Mrjie7205/serenity-bottleneck-hunter

Descubra ações upstream ignoradas mapeando cadeias de suprimento inversamente

Ver no GitHub
BomPluginScore · de 0 a 100

Skill Claude que mapeia cadeias de suprimentos ao inverso para identificar ações de gargalo ignoradas pelo mercado. Implementa a metodologia pública do trader Serenity, aplicando 9 arquétipos de gargalo com rigor de engenharia e proteção contra alucinações de LLM.

Instalação

$npx convez-ai Mrjie7205/serenity-bottleneck-hunter

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
413
Forks
71
Acompanhando
2
Saúde
42%
Último commit
há 1 mês
Versão
v0.2.0
Issues abertas
1
Contribuidores
2

Casos de uso

  • Análise de temas de investimento capex para identificar candidatos overlooked
  • Mapeamento de dependências industriais e nós críticos em cadeias de suprimento
  • Pesquisa de valor com foco em gargalos upstream antes da instituições

Veredito

Skill sofisticado e bem-engenheirado que combina análise de cadeias de suprimento com disciplinas rigorosas contra alucinações de LLM (double-check de tickers, validação de preços, auditoria de ETFs). Ideal para pesquisadores de valor experientes, não para iniciantes em investimento. O foco em candidatos antecipados (não crowded) oferece verdadeira diferença analítica.

README

A tradução para o português deste README ainda está na fila.

Serenity Bottleneck Hunter — a Claude Skill

GitHub stars GitHub forks License: MIT Claude Skill

Given an investment theme, this skill reverse-maps the supply chain to surface overlooked upstream "bottleneck" stocks — distilled from the publicly shared methodology of X/Twitter trader Serenity (@aleabitoreddit).

给定一个投资主题,复用 X 博主 Serenity 公开分享的"供应链瓶颈逆向映射"方法,独立挖出被市场忽视的上游瓶颈股(而非抄他已喊过的票)。

⚡ 30 秒装上 / Quick start

Claude Code(推荐):

# 克隆到个人 skills 目录,立即生效
git clone https://github.com/Mrjie7205/serenity-bottleneck-hunter.git ~/.claude/skills/serenity-bottleneck-hunter

然后在 Claude Code 里直接说:

用 Serenity 瓶颈猎手方法,分析「AI 数据中心电力」这个主题,给出候选标的

claude.ai / Claude Desktop:把仓库打包上传为 skill,或直接把 SKILL.md 拖进对话让 Claude 照着执行。

价格数据(可选):export EODHD_API_KEY=你的key(EODHD 全球覆盖最广,海外股推荐);没有 key 时自动回退 yfinance(美股 OK)。禁止用 WebSearch 猜价格 — 这是 skill 的硬纪律。

📊 输出长什么样 / Report anatomy

每个主题产出一份自包含单文件 HTML 报告(零依赖,浏览器直接打开,模板见 reference/report_template.html),从上到下:

区块内容
一句话结论开门见山:这个主题值不值得做、最佳切入点是什么
Step 1 · 资本开支确定性3-4 个关键数字:谁在花钱、花多少、确定性来源
30 秒看懂给非从业者的主题速览(术语友好度纪律)
Step 2 · 产业链网状图5 层逆向拆链,SVG 自动绘制依赖边,瓶颈节点双规则自动判定:漏斗型(入度≥2 出度≤1,金边)+ 枢纽型(入度≥2 出度≥2,多对多最难绕开,酒红边)
⚑ 本次行动点头条位:最多 2 张行动卡(设什么警报 / 什么条件做什么)无视排序置顶——读者 10 秒拿到本次唯一要做的事
Step 4-6 · 候选 leaderboard核心区,每候选一行:🟢候选 / 🟡观望 / 🔴排除 + 水位标尺(贴顶/高位/中位/低位/贴底 + 距高点% + 近1月/3月,人话化动量)+ trigger 分层(⚙ 可自动盯 / 👁 需人工)+ 行内三道闸小标 + 判定史(同标的历史判定:旧价→今价±%、对错复盘);原型/基本面/情景/风险折叠进"详情",按 stage 从 early→extended 排序
⭐ 跨主题信号本主题候选与其他已扫主题的交集 — 被多个 capex 周期同时锁定的 root 节点(⑤原型)
Step 5 · 三道闸门穷尽性 A / ETF audit A+ / ticker 验证 A++ 的逐项通过情况(过/半过/不过)
Step 7 · 落地结论仓位思路 + 触发器清单(什么条件下重估哪只)
底部数据备注(价格快照日期/来源)· 术语速查表(自动从正文抽取,默认折叠)· 引用链接 · 免责声明

报告中的每个 🟢🟡🔴 判定同步写入 tracking/forward_picks.csv(带日期锁定),供日后向前验证。

What it does / 做什么

Theme → reverse supply-chain map → apply 9 "bottleneck archetypes" → output a short list of overlooked upstream candidates with: thesis, archetype, valuation, entry-timing (Mode-A "buy early momentum, not the dip"), target/timeframe, and risks. The edge is being early to the theme, not chasing crowded names.

主题 → 逆向拆解供应链 → 套用 9 大瓶颈原型 → 产出被忽视的上游候选 + 论点/估值/入场时机/目标价/风险。核心是早于机构发现主题,不追拥挤标的。

🛡️ 工程化纪律 / Engineering discipline

这个 skill 的差异点不只是方法论,是把 LLM 的已知失败模式工程化拦截(每条纪律都来自一次真实翻车,教训档案写在 SKILL.md):

防线防什么工具
A 穷尽性凭记忆列候选会漏(漏过 PANW/CRWD 级主仓)已知玩家全集 audit,显式标 covered/private/acquired
A+ ETF auditLLM 的"知名度偏见"scripts/theme_etf_coverage.py 拉主题 ETF 持仓做兜底
A++ ticker 双向验证LLM hallucination 错位(把 603297 永新光学当成绿的谐波,价格/判定全反)scripts/ticker_truth.py ground-truth 库 + verify_tickers.py git pre-commit hook 自动拦截
价格纪律用 WebSearch/记忆猜价格scripts/price.py 强制 EODHD→yfinance,9 字段全留档
公司状态检查LLM 过期先验:私有→已 IPO(SpaceX 上市当天报告仍标"私有")/ 上市→被收购(SkyWater 被 IonQ 买走)判定前强制搜一次 acquisition/IPO 状态;状态断言必须带日期
向前验证事后吹回测tracking/forward_picks.csv 带日期锁定的样本外记录
防循环论证把回填的已知赢家算进"向前业绩"(曾把均α 灌水到 +90%,真实值为负)标 历史种子 的回填行强制从一切统计中剔除

Use / 用法

  • As a Claude skill: drop this folder into your skills directory (or install the .skill bundle), then ask Claude e.g. "用 Serenity 的方法分析 AI 数据中心电力 这个主题,给候选标的".
  • Or just point Claude at SKILL.md.
  • 进阶:跨主题取交集找"被多个 capex 周期同时锁定"的标的(Serenity ⑤原型),见 tracking/cross_theme_scan.py。

Structure / 结构

SKILL.md                         # 主流程:主题→挖股 7 步 + 两套择时 + 输出模板 + 教训档案
reference/
  methodology.md                 # 方法论(理念/筛选清单/两套择时/回避清单/风险)
  supply-chain-and-archetypes.md # 元框架 + 产业链速查表 + 9 大瓶颈原型库
  example_commercial_space.md    # 完整 worked example(商业航天)
  company_desc.md                # 公司业务描述库(带时间戳,90 天 freshness)
  ticker_truth.csv               # ticker ground-truth 库(L1,防 hallucination 错位)
  TICKER_HYGIENE.md              # L1+L2+L3 三层防御使用文档
scripts/
  price.py                       # 价格/动量(EODHD → yfinance 回退,严禁猜价)
  theme_etf_coverage.py          # ETF 持仓穷尽性 audit(A+ 防漏)
  ticker_truth.py                # ticker 验证/解析 API(L2)
  verify_tickers.py              # git pre-commit hook 扫描器(L3)
tracking/
  forward_picks.csv              # 向前(样本外)验证:带日期锁定的候选记录
  cross_theme_scan.py            # 跨主题节点矩阵(⭐ 跨 capex cycle 信号)
  score_tracker.py               # 事后拉价打分

Data / 数据

  • Price & timing: scripts/price.py 自动按 EODHD(EODHD_API_KEY)→ yfinance 顺序回退。EODHD 全球覆盖最广(海外股推荐);yfinance 无需 key,美股 OK 但非美股常有 gap。WebSearch 一律不用于抓价格——猜测视为流程错误。
  • Fundamentals & bottleneck judgment: web research per candidate — the skill's real edge is qualitative (is it a true single-source chokepoint?), which no data feed provides.

Validation honesty / 验证说明

The only credible test is forward / out-of-sample: see tracking/forward_picks.csv (dated, rules-locked picks) + tracking/score_tracker.py (re-pulls prices later and scores them). Any in-sample "backtest" suffers look-ahead & survivorship bias and is not a performance claim.

⚠️ forward_picks.csv 里判定含 历史种子 字样的行是回填的参照锚点(Serenity 本人的原始 call,按 2026-01-02 价格记录),不是本框架的实时判定,做任何业绩统计必须剔除——混入会造成循环论证(我们自己踩过:含种子时 🟢 均α 显示 +90%,剔除后真实向前样本为负且观察期未满)。真实向前验证基线自 2026-05-26 起算。

⚠️ Disclaimer / 免责声明

  • Not financial advice — educational / research use only. / 非投资建议,仅供研究教育。
  • This project distills a methodology from Serenity's public posts. It does not redistribute his content and is not affiliated with, sponsored by, or endorsed by him. / 本项目只提炼其公开方法论,不转载其原始内容,与本人无任何关联或背书。
  • Any "validation" inside is logic-consistency + forward tracking, NOT an audited performance record. Markets are risky; do your own research. / 文中"验证"为逻辑自洽 + 向前跟踪,非经审计的业绩。投资有风险,务必独立判断。

License / 协议

MIT (see LICENSE). Methodology credit: Serenity (@aleabitoreddit) — this is an independent, fan-made distillation of publicly shared ideas.

Ver o original no GitHub

Bom para

Relacionados