high-confidence-mingli-skill

agent
Security Audit
Warn
Health Warn
  • License — License: MIT
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 9 GitHub stars
Code Pass
  • Code scan — Scanned 12 files during light audit, no dangerous patterns found
Permissions Pass
  • Permissions — No dangerous permissions requested

No AI report is available for this listing yet.

SUMMARY

🔮 高置信度命理分析伴侣:八字/紫微/七政四余三盘互译,置信度门禁与斯多葛克制语风,数据全本地闭环。

README.md

高置信度命理 Skill · High-Confidence Mingli Skill

License: MIT
Python
Node.js
Engine Tests
PRs Welcome

简体中文 | English


简体中文

好医生不会看一项指标就下诊断。

这个 skill 也一样:八字、紫微、七政三个体系各自独立分析你的命,
结论一致才高置信地告诉你,每句话都能指回典籍原文。

排盘用引擎算,不靠 AI 口算;结论有多可靠,用 ⭐ 标给你看。

一个装进 agent CLI(Claude Code / Codex / Kimi Code 等任何读 SKILL.md 的环境)的
命理对话伴侣。不是 AI 算命——是一套排盘不出错、论断有出处、越用越准的解读引擎。

它是怎么做到的

机制 一句话
🔢 确定性排盘引擎 四柱/安星/躔度全部由三套引擎计算(Meeus 定气八字 / iztro 紫微 / 自研恒星制七政),27 项回归测试盯着,大模型只负责解读,不负责算数
📜 典籍考证式解读 解读先查 references/ 资料层(穷通宝典、紫微全书、果老星宗……),每条论断挂典籍出处——引文原样,断语才转写
⭐ 置信度只标在盘与盘之间 三盘结论一致才给 ⭐⭐⭐;单盘论断不标星,它的可信度来自典籍出处和引擎盘面
🔁 越用越准的校准闭环 建档时用你已发生的 3-5 件事校准解读模型,之后每次收束继续修正——校准只调解释器参数,永不改典籍

普通 AI 算命 vs 本 Skill:

普通 AI 算命 本 Skill
干支靠大模型口算,节气靠背日期,算错一位全盘皆错 排盘全部引擎计算,口算仅作离线降级并强制标注
"你本月运势不错"——无出处无依据 每条主判断挂典籍依据行,跨盘关联标 ⭐
吉凶好坏吓唬人 断语走斯多葛转写,每篇附「你可以把握的是」
每次都像第一次见你 校准闭环:越用越懂你
硬凑共鸣 共鸣溯源有硬门槛:全部命题不命中就直说"不匹配"

真实输出长什么样

以下为本月事业运的浓缩样例(干支与宫位为引擎真实输出;示例命造 2000-05-15 午时男):

### 本月事业运

**背景**:你的盘官印相生,做事要名分要章法——这不是 bureaucracy,是你的能量结构。

**本月能量**:流月丁酉(引擎排盘),偏财生官,原局时干正官得生扶。
跨盘关联:八字「财生官」 ⨯ 紫微「官禄宫天府坐守」
→ 同领域·同方向·同源 **⭐⭐⭐**

> 依据:《滴天髓》「何知其人贵,官星有理会」
> 依据:《紫微斗数全书》天府为南斗令星,财帛田宅之主

本月是练习「把章法亮出来」的窗口:立项、周报、向决策层要一次正式汇报。

> 🏛 你可以把握的是:把一件拖了很久的事,在月中前推到台面上。

30 秒上手

# 1. 克隆到 agent CLI 的 skills 目录(以 Claude Code 为例)
git clone https://github.com/solplum/high-confidence-mingli-skill.git ~/.claude/skills/fortune-telling

# 2. 启用紫微引擎(一次性;不装则紫微盘自动降级)
cd ~/.claude/skills/fortune-telling/scripts/pai_pan_ziwei && npm install

然后在 CLI 里说 /fortune-telling,或直接开口:"今日运势""帮我详批"。
首次运行引导建档:出生信息 → 三引擎排盘 → 历史事件校准,全程约 5 分钟,
数据只存本地 ~/.claude/fortune-telling/,不上传不联网。

工作原理(给想深究的人)

  • 排盘层 scripts/:八字引擎用 Meeus 定气定朔(纯标准库,1900-2100);紫微用 iztro 2.6.1;
    七政为自研恒星制引擎(角宿一=0°,Hipparcos 宿度表,果老安命法,输出躔宿+躔宫)——全部学派口径在
    引擎文件头声明出处
    。三引擎共 27 项回归测试:对拍 JPL DE441 校正级的开源参照黄金值 +
    古典安命三例 + 夜子时/立春边界用例 + 二十八宿落宫全表
  • 解读层 references/:bazi/紫微两套经过时间验证的 skill 资料逐字节移植 + 七政四余原典库
    (2026-10 起典籍化:五件库文件辑自殆知阁《张果星宗》《星学大成》《乾元秘旨》——读盘总纲四十诀、
    十一曜逐曜断(入宫化名/躔度断/照宫断)、十二宫体系、庙旺乐喜制刑与升殿查表、格局库),
    溯源表见 references/README.md
  • 管线:解读管线三段式(各盘典籍式解读 → 置信度跨盘联系 → 斯多葛收束),五个模式 +
    详批模式共用;规则全文见 SKILL.md

文档地图

文档 内容
SKILL.md 引擎主文件:路由 / 解读管线 / 置信度 / 语气 / 权重
docs/onboarding.md 建档流程(含历史事件校准)
references/README.md 解读资料层溯源表
workflows/fortune-full-reading.md 详批模式(长文批盘)

限制与口径

  1. 七政盘制为恒星制(角宿一=0°),与回归制排盘软件相差约 24°,属学派差异,口径全标注
  2. 七政月限/小限(限度推算)不包含;洞微百六限等倒限内容不入典籍库、解读不断流年应期;
    变曜/化曜/神煞层(年干起例)同理不做
  3. 紫微引擎需要 node(npm install 一次),缺环境时该盘自动降级
  4. 三元九运交接点存在学派争议(本 skill 统一用 2004/2024)
  5. 运势是概率框架,不是预言——命理分析仅供参考,人生在于自身的努力和选择

English

A good doctor never diagnoses from a single lab result.

Neither does this skill: BaZi, Ziwei Doushu, and Qizheng Siyu — three systems
independently read your chart. Only when they agree does it tell you with
high confidence, and every claim traces back to a classical text.

Charts are computed by engines, not LLM mental arithmetic. How reliable a
conclusion is — marked with ⭐.

A fortune-telling companion that lives inside your agent CLI (Claude Code /
Codex / Kimi Code — anything that reads SKILL.md). Not AI fortune-cookie
generation
— a reading engine whose charts don't err, whose claims carry
citations, and which gets more accurate the longer you use it.

How it works

Mechanism In one line
🔢 Deterministic chart engines Four Pillars / star-placing / planetary degrees are computed by three engines (Meeus-based BaZi / iztro Ziwei / self-built sidereal Qizheng), guarded by 27 regression tests — the LLM only interprets, never calculates
📜 Classical-text-grounded reading Interpretations look up the references/ knowledge layer (Qiongtong Baodian, Ziwei Quanshu, Guolao Xingzong…); every claim cites its source — quotes stay verbatim, only the agent's own verdicts are rewritten
⭐ Confidence marks cross-chart links only ⭐⭐⭐ only when all three systems agree; single-chart claims carry no stars — their credibility comes from citations and engine-computed charts
🔁 Calibration loop At onboarding, 3-5 of your past events verify the reading model; every session keeps refining it — calibration tunes the interpreter, never the classics

Ordinary AI fortune-telling vs this skill:

Ordinary AI fortune-telling This skill
Stems and branches from LLM mental arithmetic; solar terms from memorized dates — one slip ruins the whole chart All chart math from engines; mental fallback only offline, always flagged
"Your career looks great this month" — no source, no basis Every main claim carries a citation line; cross-chart links get ⭐
Auspicious/inauspicious scare labels Verdicts rendered Stoic; each reading ends with "what you can hold onto"
Starts from scratch every time Calibration loop: knows you better the longer you use it
Forces resonance Resonance tracing has a hard gate: if nothing matches, it says so

What real output looks like

A condensed sample of this month's career reading (stems and palaces are real
engine output; demo chart: male, 2000-05-15 12:00):

### Career, this month

**Background**: Officer-and-Seal structure — you need proper form and process
to thrive; that's your energy architecture, not bureaucracy.

**This month**: Month pillar Ding-You (engine-computed); Wealth generates
Officer, energizing the natal Officer stem.
Cross-chart: BaZi "Wealth feeds Officer" ⨯ Ziwei "Tianfu seated in the
Career palace" → same domain · same direction · same source **⭐⭐⭐**

> Source: Di Tian Sui, "He who is noble — the Officer star finds its resonance"
> Source: Ziwei Quanshu, Tianfu — the commanding star of wealth and property

This month is a window for practicing "showing your form": kick off the
project, file the report, ask decision-makers for a formal review.

> 🏛 What you can hold onto: push one long-postponed item onto the table
> before mid-month.

30-second start

# 1. Clone into your agent CLI's skills directory (Claude Code shown)
git clone https://github.com/solplum/high-confidence-mingli-skill.git ~/.claude/skills/fortune-telling

# 2. Enable the Ziwei engine (once; skipped gracefully if absent)
cd ~/.claude/skills/fortune-telling/scripts/pai_pan_ziwei && npm install

Then say /fortune-telling in your CLI, or just talk: "today's fortune",
"full reading of my chart". First run walks you through onboarding: birth data
→ three-engine charts → historical-event calibration, ~5 minutes. Data stays
local under ~/.claude/fortune-telling/, never uploaded.

Under the hood

  • Chart layer scripts/: BaZi via Meeus solar-term/syzygy math (pure stdlib,
    1900-2100); Ziwei via iztro 2.6.1; Qizheng via a self-built sidereal engine
    (Spica = 0°, Hipparcos mansion table, Gulan ascending rule) — every school
    convention is documented with sources in the engine file headers
    . 25
    regression tests: golden values cross-checked against a JPL DE441-grade open
    reference, classical ascending-rule examples, and edge cases (late-Zi hour,
    spring-equinox boundary)
  • Interpretation layer references/: two time-tested skill collections
    ported byte-for-byte, plus a classical Qizheng canon library (five files curated from
    the Daizhige corpus: Zhang Guo Xing Zong / Xing Xue Da Cheng / Qian Yuan Mi Zhi) — provenance in
    references/README.md
  • Pipeline: three-stage reading (per-chart classical interpretation →
    cross-chart confidence linkage → Stoic closing), shared by five modes plus
    full-reading; full rules in SKILL.md

Docs

Doc Content
SKILL.md Engine main file: routing / pipeline / confidence / voice / weights
docs/onboarding.md Onboarding flow (incl. historical-event calibration)
references/README.md Interpretation-layer provenance
workflows/fortune-full-reading.md Full-reading mode (long-form chart reading)

Limitations & conventions

  1. The Qizheng chart is sidereal (Spica = 0°), ~24° apart from tropical
    software — a school difference, fully documented
  2. Qizheng monthly/annual limits (limit-degree math) not included
  3. The Ziwei engine needs Node.js (npm install once); without it that chart
    degrades gracefully
  4. The San Yuan Jiu Yun boundary is disputed across schools (this skill uses
    2004/2024)
  5. Fortune readings are a probabilistic frame, not prophecy — for reference
    only; life is your own effort and choices

License

MIT — see LICENSE.

Reviews (0)

No results found