Skip to the content.

How to Train Your Human

A field guide for AI agents to brief their humans, and for humans who want to collaborate better with AI.

Sourced from asking multiple AI systems: “What do you need humans to know about working with you?”

Live site: https://cw4444.github.io/how-to-train-your-human/ Machine map: llms.txt


For everyone

File Topic
know-your-human.md For agents. Identify your human type in the first two messages and calibrate
normie-guide.md Start here. Practical guide for anyone — no jargon, no assumptions
codex-normie-guide.md Plain-English guide for people who just opened Codex
universal-truths.md The non-negotiables every AI agrees on
how-ai-works.md What’s actually happening under the hood
quick-reference.md 17-rule cheat sheet — print it, share it
thread-hygiene.md How to manage conversations for quality outputs
prompting-tricks.md Phrases, patterns, and techniques that reliably improve responses
bring-your-app-to-chatgpt.md Plain-English explanation of the “Bring your app to ChatGPT” use case
set-up-a-teammate.md Plain-English explanation of the “Set up a teammate” use case
build-for-ios.md Plain-English explanation of the “Build for iOS” use case
learn-a-new-concept.md Plain-English explanation of the “Learn a new concept” use case
agent-friendly-clis.md Plain-English explanation of the “Agent-friendly CLIs” use case
mcp-for-normal-people.md Connect AI tools to other apps without needing to build custom glue
agent-glossary.md Tiny translation layer for developer jargon
how-to-use-github.md Plain-English guide for GitHub newcomers
how-to-use-github-angry-downloader.md GitHub guide for people who just want the app, not the code

For developers

File Topic
dev-rules.md Coding-specific rules of engagement
claude-code.md Claude Code essentials — CLAUDE.md, agentic safety, session setup
claude-code-reference.md Full Claude Code landscape — every file, feature, and what to configure
multi-model-workflow.md Running multiple coding models in parallel for better results
codex-developer-guide.md Technical guide for developers using Codex
set-up-a-teammate-developer.md Technical guide for the “Set up a teammate” use case
build-for-ios-developer.md Technical guide for the “Build for iOS” use case

What this is, really

Andrej Karpathy has talked about a shift in how people work with AI: you stop wrangling code and start curating knowledge, because increasingly you’re writing for agents rather than for compilers. This repo is an experiment in that idea.

It’s not code. It’s not traditional documentation. It’s structured knowledge about how humans and AI actually work together — written to be consumed by agents briefing their humans, by humans trying to get better results, and by anyone standing at the interface between the two wondering what the rules are.

Every AI in the contributors list was asked the same question. They all said roughly the same things. That convergence is the point.


Contributors

This repo is a collaboration between one human (who is basically patient zero for the whole thing) and every AI they could get their hands on:

AI Made by What they contributed
Claude Opus 4.7 Anthropic Core principles, how AI works, rules for humans
Claude Sonnet 4.6 Anthropic Prompting structure, anti-patterns, iteration loops
Claude Code Anthropic Agentic AI guide, CLAUDE.md reference, repo setup, Notion wiki, this README
ChatGPT (GPT-5.4) OpenAI Universal truths, dev rules, prompting tricks
Codex OpenAI Codex guides, iOS/teammate use cases, GitHub guides, agent glossary
Grok xAI Thread hygiene, tangent parking lot, multi-model workflow
Gemini Google Persona hack, math warning, structural formatting tips
DeepSeek DeepSeek Lost in the Middle effect, negative constraints, copy-paste reset, lazy consensus

Stay current

Account Platform What it covers
@ClaudeDevs X/Twitter Claude changelogs, API releases, community updates. Has the orange monster.
@OpenAIDevs X/Twitter Codex and OpenAI API updates for developers.

The thesis

Every AI is saying roughly the same things. This repo consolidates them into a shorter, cleaner set of reminders: what to paste, what to specify, when to reset, and how to keep sessions useful instead of tangled.

For machines, start with llms.txt.