Mirror/liki

Liki — Professional Skill for Chinese Metaphysics
Built to professional standards: astronomical-engine charting, classically-sourced judgments, verifiable conclusions
BaZi · ZiWei · Liuyao · QiMen · Date Selection · Feng Shui · Naming

npx skills add ml8s/liki

In 30 Seconds

After installation, your AI assistant gains 4 metaphysics skills:

Skill	What you can ask	Try this
liki-bazi destiny	Marriage, career, wealth, health, education, personality, family, compatibility, full life report	`Read my BaZi, born 1990-05-20 12:00 in Beijing, male`
liki-naming naming	Baby naming, renaming, Chinese names for English speakers, name evaluation	`Name my baby, born 2024-06-10 in Guangzhou, male, surname Chen`
liki-divination divination	Liuyao (outcome & timing), QiMen (direction & decision), auspicious date selection	`Will this work out? When will I see results?`
liki-fengshui feng shui	Bazhai chart & layout, Xuankong flying stars, annual feng shui	`How is the feng shui of my home?`

What professional standards mean here:

Charts are computed by an astronomical engine (true solar time, second-level solar terms) — the AI never invents numbers
Judgments come from 756 truth-table rules, each citing classical sources (Ziping Zhenquan, Dih Tian Sui, etc.)
Independently evaluated on 160 competition questions with answer isolation

Installation

npx skills add ml8s/liki          # install all 4 skills

Install one:

npx skills add ml8s/liki --skill liki-bazi       # destiny (BaZi + ZiWei)
npx skills add ml8s/liki --skill liki-naming     # naming
npx skills add ml8s/liki --skill liki-divination # divination
npx skills add ml8s/liki --skill liki-fengshui   # feng shui

After installing, start like this:

Give me a full life reading, born 1990-05-20 12:00 in Beijing, male
Are we compatible? I was born 1992-03-15, she on 1994-08-20
How will my career and wealth go in 2026?

User Guide

Getting Started

Prepare: birth date (Gregorian), birth time (to the minute if possible), birth city, gender.

Just send it — birth info and your question in one message:

Help me look at marriage, female, born 1992-03-15 14:30 in Guangzhou

Missing details are fine: if you only know "morning" or don't know the time, the skill will follow up or start the calibration flow (see FAQ).

How to Talk to the Skill

The skill works like a practitioner — one topic at a time, with structured analysis you can drill into:

You want to…	Say
Ask why	`Why?` `What's the basis?`
Ask about a year	`What about 2026?` `Next three years?`
Switch topic	`What about wealth?` `Health?` (same chart, no re-compute)
Compatibility	`Are we compatible?` (provide both birth infos)
Full report	`Give me a full life reading`

Detailed Guide by Skill

liki-bazi (BaZi + ZiWei dual-chart)

Ask by life domain — the skill automatically charts, queries judgment tables, and gives conclusion + basis + timing:

Marriage: When to marry? Will we divorce? What's my partner like?
Career: Which industry? Startup or employment? Which years shift?
Wealth: Wealth source? Which years gain, which lose?
Health: Which organ systems? Which years to watch?
Education: How far? Exam luck?
Personality / Family: What's my personality? Parents/children affinity?

Output format: conclusion first, basis attached. Every conclusion traces to specific steps and classical sources.

liki-naming

Name my baby, born 2024-06-10 in Guangzhou, male, surname Chen

Flow: BaZi yong-shen → five-element supplement → candidate filtering → composition and evaluation → recommended names (with source and basis).

Also supports: renaming, Chinese names for English speakers, name evaluation.

liki-divination

Ask about a specific event's outcome and timing — the more specific, the better:

Liuyao: Will this project succeed? When will I see results?
QiMen: Which direction? Should I sign now?
Date selection: Best day to move / sign / open?

Output: hexagram basis → one-line verdict → timing.

liki-fengshui

Bazhai: What's my ming gua? How to arrange door/kitchen/bedroom?
Xuankong: Is my home favorable this period? 2026 annual cautions?

FAQ

Don't know the exact birth hour? Offer 2-3 candidate hours + 3-5 life events with years; the skill cross-checks and infers the most likely hour (with confidence). Babies/teens skip calibration.

Does it need internet? Chart computation goes through the liki.hk engine. If unreachable, the skill says so explicitly — never falls back to "AI guesswork".

Is my birth data stored? No. The skill explicitly commits: no birth-info storage outside your conversation, no real names requested; chart data lives only in your chat context.

How should I interpret results? Every conclusion carries its basis and classical citation — verify it yourself. Traditional cultural perspective, not medical/legal/financial advice.

How do I update? The skill self-checks its version on startup; when prompted, re-run: npx skills add ml8s/liki -y.

Why It's Trustworthy

Engine-computed, not AI-invented — charts come from a Go astronomical engine: true solar time, DST, longitude-based timezone, VSOP87D second-level solar terms. The model interprets; it never computes charts.
Sourced judgments — 41 truth tables with 756 rules, each with a classical-citation column.
Dual-system cross-check — BaZi leads, ZiWei verifies; conflicts resolved with explicit evidence.
Auditable process — every step fills a checklist; conclusions trace back to specific steps.
Independent evaluation — 160 competition questions (MingLi-Bench), answer isolation, public data (tests/).

For Developers

Architecture

skills/liki-bazi
├── SKILL.md    ← rules (process skeleton + hard constraints)
├── app/        ← process (10 cards: marriage/career/wealth/…)
├── domains/    ← knowledge (bazi 16 + ziwei 8 docs)
└── tools/      ← tools (6 Python tools + 46 judgment CSVs + 2 factor tables)
repo root
├── engine/     ← Go JSON-RPC astronomical engine (8 domains)
├── tests/      ← evaluation (160 grouped cases + answer isolation)
└── scripts/    ← build / distribution index

Call chain: SKILL.md routes to an app card → the card calls the six Python tools (agent_cli.py orchestrates RPC charting, factor evaluation, and CSV matching) → interpreted via domain knowledge → rendered by the card template. RPC methods are invisible to the liki-bazi LLM.

Engine Image

Images are auto-published on GitHub Releases (CI full tests → build + push + smoke test): docker pull ghcr.io/ml8s/liki-engine:latest. Build from source: cd engine && docker compose -f deploy/docker-compose.yml up -d --build.

Quick Start

make hooks         # install git hooks (once)
make test-all      # full: skills unit + engine (lint/vet/race/integration/smoke) + e2e
make check         # table schema + doc contracts + version consistency
make build-archive # pack 4 skills + generate the distribution index/archive digest

Design Principles

Single responsibility per layer: root=rules, app=process, domains=knowledge, tools=tools
Single data source: LLM tool contracts from tools/skill-tools.json; judgments from CSV truth tables
Dual-system: BaZi leads, ZiWei reviews, conflicts explicit
CalVer (date-stamped VERSION + CHANGELOG; startup version self-check; tags on milestones)
Evaluation-driven: independent grading, answer isolation, public data

See CONTRIBUTING.md and CHANGELOG.md. Design references include mingli-skills, bazi-skill, iztro, and MingLi-Bench.

See docs/FACTOR_MODEL.md for liki-bazi's factor contract and full factor inventory.

License & Disclaimer

MIT. Conclusions are from a traditional cultural perspective, for research and reference only — they do not constitute medical diagnosis, legal advice, financial forecasts, or major life decisions. Please stay rational and proactive.