hsb.horse
I built a skill that analyzes a user’s personality from chats with AI agents such as Claude and outputs HEXACO and MBTI scores in YAML format.
As a bonus, I also added a feature that picks three Pokemon “similar to you” based on the HEXACO scores.
Motivation
At the end of last year, prompts like “From all of my 2025 conversation logs you remember, sharply describe my biggest strengths and the issues I was unconsciously avoiding from a psychological perspective” were popular. Honestly, I did not find the results trustworthy.
In technical conversations or when I ask about areas I do not already know well, I tend to make sure the first prompt contains the necessary information because I am an engineer. As a result, the feedback kept collapsing into things like “metacognition” or “abstraction/structuring.”
I have been thinking that conversations with AI reveal a lot about how a person thinks and what they value. If that could be quantified, it seemed interesting.
I chose HEXACO because it has more factors than the Big Five, and especially because Honesty-Humility stands as an independent dimension. MBTI is there mostly as a reference.
How to use it
Just say “analyze my personality” in the middle of a chat.
You can also specify a language.
analyze my personality lang:enDesign notes
credibilityis calculated from the amount of speech and the diversity of content on a 0-100 scale- If the chat is mostly technical questions, personality traits are harder to observe, so
credibilitygoes down - It does not read past chats. It only looks at “how you behaved inside this chat”
Impression
The skill itself is simple and fits inside a single SKILL.md. No scripts or assets were needed.
Accuracy is honestly not something I would fully trust, but it may still help people look at their own conversation patterns more objectively.
Summary
I tried building a skill that analyzes personality traits from conversations with AI.
If the session is only technical, the result tends to be biased, but as a tool to look back on your own behavior on a per-session basis, it is still interesting.
The complete implementation is published in this Gist.