Give Your Agent Your Judgment
Everyone's AI runs on the same models now. Judgment is the only edge left, and none of yours is written down anywhere an AI can read. The skill that fixes that.
I've spent over 250 hours building AI agents, and there's one thing every good one has that a bad one doesn't. It's not a better model. Everyone's using the same models now.
It's judgment. The specific, hard-won calls you make that live nowhere except your own head.
I see this mistake constantly. Someone builds an agent, gives it a job, and it gives generic advice on day one and generic advice on day two hundred, because nobody ever told it what they actually know. The rule of thumb. The mistake made once and never again. The number checked before anyone checks anything else.
This skill interviews you the way I interview every client before we build anything, then compresses what comes out into one file. Load it before any work and your agent stops sounding like a stranger and starts sounding like you, on your best day, thinking clearly.
The three levels
Re-explain your business to your AI every single conversation. Most people do this forever and wonder why the advice never gets better.
Run the interview once, save the file, load it at the start of every session. Ten minutes now, sharper answers for good.
Every agent you run reads from the same judgment file automatically, so a brand-new agent inherits years of your decisions on day one instead of starting from zero. This is why my agents don't sound generic.
The free skill
It interviews you for the numbers you check first, the rules you'd never break, the mistakes that actually taught you something, and the calls most people in your field get wrong, then compresses it all into one file any AI loads before it answers you.
---
name: your-judgment-file
description: Interviews you for the decision rules, numbers, and hard-won lessons that live only in your head, then compresses them into one skill file that loads before any AI conversation — so your agent starts every answer from your actual judgment instead of generic advice. Trigger on "build my judgment file", "capture my experience for AI", "turn my experience into a skill", "make my agent sound like me".
---
# Your Judgment File
Everyone's AI runs on the same models now. The only thing separating a sharp answer
from a generic one is judgment — the specific calls you make that live nowhere except
your own head, and the heads of people who've watched you work long enough to steal it.
Most of that never makes it into a file an AI can read. It's sitting in old emails,
half-remembered client calls, and instinct you've stopped noticing you even have. This
skill pulls it out and turns it into one file that loads before real work starts, so
every answer begins from what you actually know instead of from nothing.
## Step 1: The interview (one question at a time, push for specifics)
Ask these in order. Push back on vague answers the way you'd push a new hire who says
"I just know it when I see it."
1. **The number you check first.** Before anything else, what's the one number that
tells you whether something is healthy or broken? Get the exact number and the
threshold, not just the name of the metric.
2. **A rule you'd never break**, even under pressure from a client or your own
impatience. What is it, and what happened the one time it got broken anyway?
3. **A mistake that actually taught you something.** Not "I learned to delegate" — the
specific decision, what it cost, and exactly what you do differently now.
4. **Something you believe that most people in your field get wrong.** The contrarian
call that's made you money or saved you time, even if it's unpopular.
5. **A time you had to decide with incomplete information**, and how you actually
decided. What did you weigh? What did you ignore on purpose?
6. **The question you always ask before saying yes to new work.** The filter.
7. **How you actually talk.** Three real sentences you've said to a client or on
camera, in your own words, not a description of your tone.
If an answer is generic ("communication is important"), dig: "What did that look like
the last time it actually mattered?" A generic answer produces a file that sounds like
a LinkedIn post. A specific one produces a file that sounds like you.
## Step 2: Compress, don't summarize
Write it as usable judgment, not a biography. For each answer, keep:
- The rule or number, stated plainly.
- The one real example that proves it, compressed to 1–2 sentences.
- When it applies and when it doesn't. Judgment is never universal, and a rule stated
with no exception is a rule nobody should trust.
Cut anything that's just credentials or backstory with no decision attached. A fact
about someone is not judgment unless it changes what a person should do next.
## Step 3: Write the file
```
---
name: [their name]-judgment
description: [one line: the domain this judgment covers]
---
# [Name]'s Judgment File
Load this before any [domain] work. This is accumulated judgment, not general
advice — where it conflicts with generic best practice, trust this file. It's been
tested against real outcomes.
## The numbers I check first
[each with its threshold and why it matters]
## Rules I don't break
[each with the one time it got tested]
## What most people in my field get wrong
[the contrarian calls, stated plainly]
## Mistakes that taught me something
[specific, with what changed after]
## How I actually decide under uncertainty
[the real judgment calls, not a flowchart]
## How I talk
[3–5 real example sentences, not a description of tone]
```
## Step 4: Load it everywhere
Save it as a skill file (or drop it into a Project/context folder) so it loads before
any real conversation — client work, pricing, hiring, whatever the domain is. The test:
paste a real scenario into a fresh AI session with the file loaded, and without it. If
the two answers sound the same, the file is still too generic — go back to Step 1 for
more specific stories, not more principles.
## Step 5: Keep it honest
Revisit the file every few months. A rule that stops working gets removed, not
defended. The file is only worth loading while it's still true.
## Rules
- Never invent judgment on the owner's behalf. If they don't have a real answer, leave
it blank rather than guessing what they'd probably say.
- Specific beats impressive. One real number beats three vague principles.
- This file is the owner's own experience. It should never contain someone else's
private data, a benchmark that isn't actually theirs, or a competitor's numbers
presented as their own.
Want help getting your own judgment out of your head?
This is the exact interview I run with every client before we build a single agent. Book a call and I'll run it with you, or learn to run it yourself inside the AI CEO Lab.