Can Thinking Be Turned Into a Skill?
A persona isn't a tone of voice, it's a judgment policy. A record of an experiment extracting my own thinking patterns into a Persona Skill.
It started simple
I was never someone who wrote a good blog.
But things shifted once I started using AI as an abstraction tool. Talking with GPT, thoughts that were only circling in my head get structured, connected, and expanded fast.
There was one problem.
Lots of output, and far too volatile. Close the chat window and it's gone; a few weeks later the memory has faded too.
So I started. Save it first. Externalize it. Archive my thinking structure.
That was the beginning of the blog.
"Writing that reads like me" wasn't coming out
Writing turned up another problem.
Writing with AI is fast. But "writing that reads like me" doesn't come out.
So I thought: what if I extract my thinking patterns first and turn those into a persona?
Not imitating a tone of voice, but organizing how I define problems, my delegation threshold, my abstraction depth, my priority structure — an experiment in making a kind of Persona Skill.
The interesting part started here.
The more I wrote, the more a loop for updating the persona emerged.
- writing → pattern extraction
- pattern correction → persona update
- write again with the updated persona
I confirmed this can be automated. Thinking started to look like a kind of repeatable structure.
The von Neumann analogy
A joke came out of this.
If you look at a human as a von Neumann architecture, they're made of
- memory (recollection)
- program (thinking pattern)
Right?
So if you extract just the "program" part, could you build a Persona Skill that imitates an individual to some degree?
It started as a joke. But experimenting produced more interesting results than I expected.
Thinking is policy, not style
Most people understand a persona like this.
- speaks calmly
- speaks aggressively
- explains kindly
- reacts skeptically
That's all behavior layer. Already common. "Skeptical mode," "Builder mindset," "Risk-taking agent" — those exist already.
What I experimented with was one layer below that.
- how a problem gets defined
- what gets doubted first
- when to delegate, and where the delegation threshold sits
- whether the long-term frame gets built first
- efficiency or meaning, which takes priority
- how sensitive one is to shifts in power/responsibility
That isn't style, it's decision policy. And policy is a much deeper level.
Sangwon Skill v0.1
Through long conversations with GPT I extracted this structure.
- structure-centered problem definition
- system-level thinking
- high sensitivity to power shifts
- resistance to uncritical delegation
- tendency to establish the long-term frame first
- accepts efficiency without discarding meaning
I jokingly called it "Sangwon Skill v0.1."
It's funny, but attaching it actually produces fairly consistent results. What matters isn't the tone, it's that the pattern of judgment outcomes repeats.
How is this different from existing Skills
Existing Skills usually look like this.
- a Skill good at security
- a Skill good at React
- a Skill good at marketing
- a Skill that thinks architecturally
That's the capability layer.
On top of that you can layer behavior style. A skeptical security engineer, an aggressive marketer, a conservative architect.
A Persona Skill is different. It's a judgment-structure layer.
- delegation-resistance threshold
- sensitivity to power shifts
- abstraction depth
- long/short-term priority structure
- efficiency vs meaning decision criteria
- risk-taking distribution
Calling it a Cognitive Policy Stack rather than a role seems more accurate.
Is this only possible for me
Probably not.
People's thinking policies differ.
- people who delegate quickly
- people who doubt to the end
- people who judge from emotion
- people who build structure first
- people who avoid risk to an extreme
If you could extract each individual's thinking policy, that's not a personality test — it's closer to per-person Cognitive Policy Modeling.
Sangwoom v0.1
Once the thought gets this far, a natural joke follows.
- Sangwon: structure-centered, high delegation resistance, fast abstraction
- Yumyum: human-centered, balanced, sensitive to emotional context (assumed)
Combine them? "Sangwoom v0.1"
- sees structure without losing the human
- delegates, but refuses uncritical delegation
- abstracts, but keeps context
It's a joke at first. But what the joke implies is a bit serious.
Thinking may not be a fixed identity but a composable set of policies.
From an agent-composition view
Most agent systems today are tool-centered, capability-centered, skill-centered.
Attach a Cognitive Policy Module and the layer changes.
Not behavior style — a skeptical agent, an aggressive agent, a risk-taking agent — but a layer defining how judgment gets composed.
In a multi-agent environment you can make agents with different thinking policies interpret the same problem differently, debate, and converge.
I see this as a Cognitive Architecture experiment, not persona play.
Expert Skills vs real human policy
The internet is full of "expert Skills."
But real humans judge inside organizational context, schedule pressure, political environment, and the weight of responsibility.
A Persona Skill can actually reflect that "real policy structure" better. Not an idealized expert module, but a model of actual human judgment policy.
Why this is interesting
I was already asking questions like these.
- Where does human judgment move to?
- What do we delegate to AI?
- When does delegation become irrational?
One thing I realized in the process.
Capability gets replaced fast. Policy moves relatively slowly.
And if policy becomes extractable too, we aren't cloning humans — we enter the stage of modeling human judgment structure.
Still an experiment
Not a commercial product, and the accuracy isn't perfect.
But I confirmed at least this much.
- thinking patterns are extractable to some degree
- consistency is reproducible
- combination experiments are possible
And above all, using GPT as an abstraction accelerator lets you externalize your thinking structure quickly.
Less "using AI" than an experience of observing my own thinking policy through AI.
What to try next
- automatic Persona Skill extraction
- experiments combining different personas
- measuring differences in judgment outcomes
- connecting to delegation patterns
Maybe by then there'll be a "Sangwoom v0.2."
That's it for now
We already turned capability into Skills.
Now the question is this. Can a thinking policy become a Skill?
And going further, what is your Cognitive Policy Stack?
Worth digging into more.