I always thought I was born in the Zi hour.

That is what my family said: “late at night.” So every BaZi chart I had ever seen used Zi. Some readings fit, some did not. I assumed the chart was just complicated.

Then I tried a skill that refused to cast the chart until it had checked the hour.

The first useful thing it did was ask questions

I gave it my birth date, city, and gender, and said I was not sure between Zi and Chou.

It did not pick one and move on. It asked me about a few real events: when I married, when I changed jobs, when I moved, when I had a child. Then it cast both candidate charts and compared them against those events.

It told me that the Zi chart did not line up with several key years, while the Chou chart fit better. It did not declare “you are Chou” as a fact. It showed the candidates, the comparison, and the uncertainty.

That was the first time an AI astrology tool had treated my input as something worth checking.

One hour changed the whole chart

When I switched to Chou, the hour pillar changed. The relationships in the chart changed. The timing of later years changed. Some things that had never made sense suddenly had a place.

I am not saying it became “accurate.” I am saying the chart finally started from a better premise.

If the hour is wrong, everything after it is built on the wrong foundation. Most tools skip that step. This one did not.

Naming became a conversation, not a vending machine

I had a second test: naming.

The usual experience is: send birth details, wait, receive a list of names. You do not like them, but you do not know how to ask for changes. You also do not know why a character was chosen.

This skill started by asking what I actually wanted:

  • Do you need to keep the generational character?
  • Are there characters you must avoid?
  • Do you want something classical or everyday?
  • Should we avoid awkward pronunciations in your dialect?
  • What matters more: meaning, sound, or the five-element balance?

Then it looked at what the chart needed, proposed characters, and explained each choice: why this character, what it balances, how it sounds, and where it connects to the chart.

I said “too literary, make it more everyday.” It changed direction. I said “I don’t like this character.” It re-evaluated instead of shuffling the same list.

I could ask: “If we must keep this character, what else works with it?” “Does this name sound strange in my dialect?” “My family insists on this character — can we make it work?”

It kept going.

That is a different product experience. It is closer to working with a designer than buying a one-shot list.

Other things I tried

I also used it for compatibility, annual timing, a specific divination question, date selection, and a quick Feng Shui question.

What stood out was not that it could answer everything. It was that I could keep asking.

For compatibility, it did not stop at “compatible” or “not compatible.” It broke down where two people are likely to clash, which years need care, and how to communicate better.

For a job-change question, it gave a time window instead of a yes/no.

For a specific offer, it cast a hexagram, identified the key point, and gave a rough timing.

For date selection, it filtered a range and explained why one day worked and another did not.

I did not need to learn a full vocabulary before I could ask a useful question.

What I liked as a product

Three things stood out.

1. It starts with input, not output. Most AI readings start with a confident answer. This one starts by checking whether the input is good enough to answer.

2. It lets me ask why. I can ask where a conclusion came from, why this hour and not that one, why this character and not another. It explains instead of asking me to trust it.

3. It is willing to say “not enough evidence.” When the hour cannot be resolved, it says so. When the input is incomplete, it asks. That is rare in this category.

It also keeps birth information in the current session and does not write it into a profile. For a topic this personal, that matters.

What it does not do

It needs a network connection. It asks follow-up questions, so it is not a one-shot “tell me my future” tool. Some conclusions still require interpretation. And it will not invent an answer when the rules do not cover a case.

If you want a single confident sentence, it will probably frustrate you.

If you want to participate in the process, it is a different experience.

Install

The skill is open source and installs with one command:

npx skills add ml8s/liki 

It works in Claude Code, Codex, Cursor, WorkBuddy, and any MCP host.

I work on Liki Skill, so take this as a builder's user note rather than a neutral review. The parts I would keep even if the product changed are the same: check the input first, show the evidence, and let the user keep asking.

This is a traditional practice perspective, for reference only, and not professional advice.