Skip to content
Projects

ml8s/liki has 103 GitHub stars and 15 forks. GitHot measured 0 stars in 24 hours. Last checked 2026-11-16 19:53 UTC. Last push 2026-11-15.

How these counts are measured

Measured star growth

in 24h
0
in 7 days
Collecting history
in 15 days
Collecting history
in 1 month
Collecting history

Net star changes over overlapping periods, not daily totals. Missing measurements are left blank. Stars measure attention, not software quality.

About

Liki Skill — Professional Astrology Skill for AI assistants. BaZi, Zi Wei Dou Shu, Liu Yao, QiMen, Feng Shui, date selection and Chinese naming, powered by a deterministic engine with traceable, verifiable readings. Works in Claude, Codex, Cursor and any MCP host. Free, MIT.

AI

README.en.md

Original

Liki Skill logo

Liki Skill

Professional Astrology Skill.
For Astrology, use Liki.
Reproducible charts · Traceable readings · Honest boundaries
Bazi · Ziwei · Liuyao · QiMen · Date Selection · Feng Shui · Naming

中文 CI license website

Why Liki Skill

Ask a general-purpose LLM about the same chart three times and you may get three different answers. Chinese metaphysics amplifies small errors: miss a solar-term boundary by a day and the year pillar changes — every reading after it shifts.

So Liki Skill turns each error-prone step into an explicit process: chart casting cross-checked against multiple implementations, readings emitted from rule truth tables, birth time normalized and calibrated before any judgment. Conclusions are computed, not improvised.

  • ✅ Reproducible charts — same input, same chart, every time; solar terms and true solar time are handled by the astronomical engine
  • ✅ Traceable readings — every conclusion cites its factor evidence, all the way to the classics (《三命通会》《滴天髓》…)
  • ✅ Honest boundaries — time-sensitive charts get calibrated; insufficient evidence is labeled; nothing is silently defaulted

Three deep dives

Three places most tools stay vague — and where Liki Skill differs the most.

1|The chart engine was cross-checked against three open-source implementations

The engine was not written once and shipped. 192 solar-term boundary cases (2024–2027, 24 terms per year × 2 anchors per term) are cross-checked with lunar-python, sxtwl, and bazi-calculator under a single policy: all_oracles_must_agree — one mismatch and the gate fails. The Ziwei implementation aligns with iztro-compatible data.

Where the engine disagreed with the classics, it was corrected against the original text and pinned as regression anchors:

《穷通宝鉴》 correction example: 乙木向阳,专取丙火;水多以戊佐。 The engine output matches this wording exactly, alongside full alignment with a 120-entry classical reference table.

2|Direct readings from a truth table fitted with exam questions and classics

Readings are not generated by the model — they are hit from a truth table: 829 assertions, 893 condition groups, 1,355 condition rows, driven by 511 natal factors and 111 annual factors. A hit yields a direct conclusion with its sources; a miss is reported as a miss. There is no room for improvised wording. Every assertion carries its evidence and classical source — for example, this real row:

Assertion Basis
Five elements lack Wood: liver / gallbladder / sinews / eyes run weak Wood governs the liver (《黄帝内经》 "East generates wind; wind generates wood; wood generates sour; sour generates the liver")

Two design details:

  • Annual assertions must carry year-activated factors, or data validation fails — preventing natal traits from being repeated every year;
  • 160 professional competition questions (Mingli160) are used for rule-chain coverage and zero-hit checks, as a manual calibration/evaluation set.

3|Time calibration — even when you hold a birth certificate

The most error-prone step in chart casting is "determining the time":

  • Hospital clocks record standard time, not the sun at your birth moment — depending on longitude, true solar time can drift by half a shichen (double-hour);
  • Born near a solar-term boundary, a few minutes change the year pillar;
  • After 23:00, the "late Zi" rule makes the day pillar assignment error-prone.

So Liki Skill makes calibration an explicit process instead of silently casting your chart from wall-clock time:

your time + birthplace → normalization → candidate set (each with reasons)
→ full chart per candidate → hour-pillar-sensitive facts (pillar / ten gods / shen sha)
→ compare against your real events (2-3 candidates, 3-5 events with years) → calibrated result + evidence coverage
→ insufficient evidence? labeled as such — never silently defaulted

Up and running in 30 seconds

Step 1: install the skill (one line)

npx skills add ml8s/liki

Step 2: just ask

Goal Ask
Destiny Calculate Bazi for a male born 1990-05-20 12:00 in Beijing.
Naming Name a boy born 2024-06-10 in Guangzhou; family name Chen.
Divination Can this project succeed? When will I know?
Date selection Which day next month is good for moving?
Feng shui How is my home's feng shui?

For destiny readings, provide the birth date, exact time when possible, birth city, and gender. If data is incomplete, the Skill asks follow-up questions or enters time calibration. When prompted, update with npx skills add ml8s/liki -y.

Ask an LLM directly, or use Liki Skill?

Ask an LLM directly Liki Skill
Charts Mental math by the model; solar terms / true solar time are error-prone; results vary Astronomical engine; same input, same chart, every time
Readings Freeform, no sources 829 rule truth-table assertions, each with classical sources
Time & boundaries Wall-clock time used as-is; boundaries silently resolved Candidate set + reasons → per-candidate calibration; insufficient evidence is labeled
Output Plain text, hard to build on Structured JSON + SHA-256 digests; plug it into your product
Verification None Open source; the whole chain is auditable

Distribution variants

Liki Skill builds two distribution variants from the same methodology:

Variant MCP access Payment
liki-free Direct engine-mcp and counsel-mcp None
liki-pro Through the liki-mcp paid product MCP Supports SkillPay / 402 payment flow

The free source installs with npx skills add ml8s/liki. Paid packages are distributed by channel to SkillHub / WorkBuddy.

Trust and boundaries

Three pillars of reliability

Pillar Fact
Traceable (competence) Every reading → hit factors → factor basis → classical text: a four-layer traceable chain
Reproducible (reliability) Engine cross-checked against mainstream implementations; key outputs carry SHA-256 digests
Bounded (low self-orientation) Birth data stays in the session; no real names, no profiles, no feedback payloads

Engineering and privacy facts

  • Explicit dual-system review: Bazi and Ziwei are computed separately; conflicts are listed by evidence layer.
  • Skills and services do not compare versions; unknown tools return an explicit error.
  • Images are built locally and deployed by commit sha.
  • Sensitive questions go through a support-first safety policy.
  • City lookup uses only the built-in WGS-84 table and never makes an external geocoding request.

Honest disclaimers: boundaries are part of reliability

  • Mingli160 is a coverage and zero-hit check, not an independent accuracy benchmark or proof of correctness.
  • Conclusions are conditional interpretations from a traditional cultural perspective. They are not medical, legal, investment, or major life advice.

Who it is for

  • People who want to understand their chart seriously — reliable casting, checkable sources, no need to parse classical texts yourself.
  • Practitioners and content creators — a consistency tool: the same chart reads the same, for anyone, at any time.
  • Developers — standard MCP tools with structured JSON output, ready to embed in your product.

FAQ

What if I don't know the exact birth hour?

The Skill asks follow-up questions or uses real events for calibration. Insufficient evidence is labeled "insufficient calibration evidence"; the hour is never silently defaulted. Naming skips calibration and uses an expected-element strategy; babies and teenagers are read directly from the provided hour.

I know my exact birth time — why calibrate?

Clock time is not destiny time: standard time (say, Beijing time) can drift from your birthplace's true solar time by half a shichen; a few minutes near a solar-term boundary changes the year pillar; the late-Zi rule makes the day pillar error-prone. These errors hide at boundaries you cannot see — so even with an accurate clock time, Liki Skill normalizes first: list candidates with reasons, cast each, then calibrate against your real events. It is the first step of chart correctness.

Does it need internet?

Yes. The Skill talks to the hosted engine-mcp (charts) and counsel-mcp (readings) MCP endpoints — install and ask, no configuration needed.

Is my birth data stored?

No. Birth data remains in the current conversation context. It is not written to a local profile or submitted through feedback.

How do I update?

Run npx skills add ml8s/liki -y when prompted. The Skill fails closed instead of calling stale RPCs.

Repositories and deployment

Four repositories: liki, liki-agent-runtime, liki-web, liki-deploy. Images are built locally (tag = commit sha); liki-deploy runs the full-system e2e, deploy, and smoke.

License and disclaimer

Code is MIT licensed. Conclusions are traditional cultural interpretations for reference only. They are not medical, legal, investment, or major life advice. The bundled geographic data is not relicensed under MIT; see the NOTICE.

Full usage: user guide; contributing: CONTRIBUTING; release history: CHANGELOG.

For Astrology, use Liki.