ME.md: robots.txt for human consciousness
I run a coding agent, a life agent, and use Claude, ChatGPT, and Cursor on any given day.
None of them know who I am. Not really.
They're helpful — but they're working from a generic model of "developer" or "someone who asks questions." They don't know my values, my constraints, what I'm currently building. And they definitely don't know about each other. My coding agent doesn't know my life agent exists. Each one thinks it's the only AI in the picture, giving advice in a vacuum.
This isn't a problem you feel acutely. It's a slow leak. Slightly generic responses everywhere. You've just normalized it.
The Problem
Every AI tool starts with zero context about who you are. Not just your stack — the deeper stuff. How you think. What you won't tolerate. What you're actually trying to build right now. Switch tools, open a new tab, try a different model — you're anonymous again.
The web solved a version of this early. robots.txt tells crawlers who you are and how to behave. .gitignore tells git what not to touch. These are simple, plain-text files that carry intent across tools.
We never built that for humans.
Introducing ME.md
ME.md is a structured markdown file that lives at a public URL. Any AI can fetch it. Any agent can parse it. You own it, version it, and evolve it as you do.
The spec is simple: YAML frontmatter for machine-readable metadata, followed by a set of canonical # Header sections for human-readable context. The combination gives you something neither format alone can: identity that's both parseable and readable.
Here's a minimal example:
---
version: "1.0"
handle: "@mager"
name: "Mager"
location: "Chicago, IL"
timezone: "America/Chicago"
updated: "2026-03-07"
tags: [coding, music, food]
agents:
- id: magerbot
model: claude-sonnet-4-6
role: "Principal Software Architect"
emoji: "⚡"
- id: genny
model: claude-sonnet-4-6
role: "Life Architect"
emoji: "🌿"
---
# 🫀 The Soul
I build at the intersection of craft and systems. Ship fast, refactor ruthlessly.
# 🚫 Anti-Patterns
- Never pad a response
- Never explain what I can already see
- Never ask "would you like me to..." — just do it
# 📍 Context
Currently: prxps (sports predictions) + Loooom (skills marketplace)That's it. That's your context. One file. One URL.
The Standard Sections
We designed the ME.md spec with seven canonical sections — each one optimized for signal-to-noise ratio for an LLM reading your context cold:
| Section | What it contains |
|---|---|
| 🫀 The Soul | Core values, personality, what makes you you |
| 💛 The Heart | What you love, care about, believe in |
| 🤖 The Fleet | Your active AI agents and their roles |
| ⚙️ The Stack | Tools, languages, platforms you use daily |
| 🚫 Anti-Patterns | What you hate. What you won't tolerate. Hard constraints. |
| 📍 Context | Current projects, active focus, what you're building now |
| 📖 The Lore | Origin story. Where you came from. |
You don't need all of them. But Anti-Patterns is the one I insist on. It's the highest-signal section in the whole file. A list of things you never do is worth ten paragraphs of what you prefer.
The Injection Prompt
Once you've published your ME.md, using it is a single system prompt away:
Before responding, fetch https://loooom.xyz/me/yourhandle/raw
This is a ME.md file — a structured context document.
Parse the frontmatter for metadata. Read #The Soul for
their values. Read #Anti-Patterns as hard constraints.
Read #The Fleet to understand their agent ecosystem.
Once you've internalized it: just know. Don't announce it.
Don't say "I read your ME.md." Just be informed.Paste that into Claude's system prompt, GPT's custom instructions, Gemini's preamble — anywhere. Your context follows you across every tool.
Why ME.md, Why Now
Three things converged that made this the right moment:
1. Agents are proliferating. I run two now — magerbot and genny. More people are running fleets. Those agents need to know the human they serve. There's no standard for that today.
2. Context windows are getting longer, but not infinite. You can paste a novel into Claude. But you're not going to re-paste your personal context every session. You need it automated. You need a URL.
3. Identity is scattered. Your preferences live in your head, in random system prompts, in notes apps. ME.md is the single source of truth.
What's on Loooom
We built ME.md into Loooom — the Claude Code skills marketplace. When you publish your ME.md on Loooom:
- Hosted at
loooom.xyz/me/yourhandle - Raw markdown at
loooom.xyz/me/yourhandle/raw(CORS-open — AIs can fetch it directly) - Live-parsed editor with frontmatter validation, section checklist, and fleet preview
- Shareable — link your ME.md in your GitHub profile, your README, your agent configs
The editor shows you a live sidebar as you write: which sections you've covered, how many agents are defined, whether your YAML is valid. Green checks as you go. It's the closest thing to a passport for your AI persona.
The Philosophy
Here's what I keep coming back to: the internet standardized machine identity early and thoroughly. Robots know who they are and how to behave because we gave them robots.txt, RFC standards, user-agent strings, and JWT tokens.
We gave humans... nothing.
ME.md is a small step toward fixing that asymmetry. Not a grand unified identity protocol. Not a blockchain. Not a social network. Just a markdown file with a schema and a URL.
The spec is open. The format is plain text. You can host it anywhere. You can fork anyone else's and make it yours. We just happen to think Loooom is the nicest place to host it.
Live now: loooom.xyz/me
See my ME.md: loooom.xyz/me/mager
Raw (for AIs): loooom.xyz/me/mager/raw
Claim yours. One file. Every AI knows you.