Identity

One markdown file that encodes who you are — your thinking patterns, judgment heuristics, epistemic guardrails, and operating principles. AI becomes dramatically more effective when it knows how you think.

1 file · 5 min setup · shared/context.md

The problem

Every AI conversation starts from zero. The model has no idea how you reason, what you value, or what quality means to you. So you get generic output. You correct it, and the correction disappears next session.

Most people solve this by writing longer prompts each time. That doesn't scale. Identity is the structural fix: a persistent file that every agent reads before producing anything.

What gets encoded

The identity file isn't a bio. It's a thinking model. Agent Prime's context.md captures five layers of who you are:

Career Arc

Your trajectory and what makes your perspective unusual. The cross-domain experience that shapes your pattern recognition.

Operating Philosophy

Your core drives and values. In Agent Prime, this is the Four Wills — Know, Think, Act, Will.

Reasoning Operations

How you actually think. Layer thinking, isomorphic mapping, adversarial self-critique, hidden dependency hunting.

Intellectual Range

The cross-disciplinary models you draw from. Chaos theory, quantum mechanics, evolutionary dynamics, philosophy — the full substrate of your pattern recognition.

Epistemic Guardrails

Nine failure modes that corrupt thinking when using AI. Premature coherence, analogy capture, forced synthesis, deference drift, and more.

Real example

Here's an excerpt from Agent Prime's actual identity file — the reasoning operations section:

shared/context.md — Reasoning Operations

1. Layer thinking.
   Any complex system has layers. The interesting failures
   and opportunities happen at the interfaces between layers.
   First question in any new domain: "What are the layers
   and where are the hidden dependencies between them?"

2. Isomorphic mapping.
   Structural parallels between systems in completely
   different domains — the same shape, different substrate.
   The bar: if you can't state the isomorphism in one
   sentence, the analogy isn't precise enough.

3. Adversarial self-critique.
   "Now rip this entire thinking to shreds." Genuine
   attempts to break your own reasoning — not devil's
   advocacy as performance.

Every agent reads this before producing output. The result: output that thinks the way you think, catches its own mistakes, and uses your vocabulary — from the first message.

How to set it up

Create one file: shared/context.md. Write down:

## Who Am I?
Your role, trajectory, and what makes your lens unusual.

## How I Think
Your reasoning patterns. What do you do when you
encounter a new problem? What's your first instinct?

## What I Value
Your quality bar. What does "good" look like to you?
What makes you reject output?

## Failure Modes
How does your own thinking break? Where do you
get lazy, overconfident, or sloppy?

Start with a paragraph for each. You'll expand it over time as you notice patterns in your own corrections. The Memory layer captures those patterns automatically.

What you get

shared/context.md

Your complete thinking model. Every agent reads this first.

~300 lines · grows over time