karpathy

LLM Wiki

A pattern for building personal knowledge bases using LLMs.

This is an idea file, it is designed to be copy pasted to your own LLM Agent (e.g. OpenAI Codex, Claude Code, OpenCode / Pi, or etc.). Its goal is to communicate the high level idea, but your agent will build out the specifics in collaboration with you.

The core idea

Most people's experience with LLMs and documents looks like RAG: you upload a collection of files, the LLM retrieves relevant chunks at query time, and generates an answer. This works, but the LLM is rediscovering knowledge from scratch on every question. There's no accumulation. Ask a subtle question that requires synthesizing five documents, and the LLM has to find and piece together the relevant fragments every time. Nothing is built up. NotebookLM, ChatGPT file uploads, and most RAG systems work this way.

imaman

name	regression-dog
description	Review code changes for regressions.

Review code changes for behavioral differences between the before and after code.

Important:

• Do NOT run tests, typechecks, linters, or build commands. CI already handles those. Focus your context budget entirely on reasoning about the code changes and its implications on logic/behavior/data/etc.

ivanfioravanti

Build a pixel-perfect Space Invaders arcade game in a single self-contained HTML file with inline CSS and JS. Replicate the original 1978 Taito arcade cabinet experience with full fidelity to the original rules, sprite designs, and mechanics.

---

CANVAS & DISPLAY

• 800×600px canvas (or scaled 2× from original 224×256 logical resolution)
• Black background, phosphor-green or white pixel aesthetic

mberman84

OpenClaw Prompts - Build Your Own AI Assistant

Prompts to recreate each piece of the OpenClaw system. Use these with any AI coding assistant.

---

1. Personal CRM "Build a personal CRM that automatically scans my Gmail and Google Calendar to discover contacts from the past year. Store them in a SQLite database with vector embeddings so I can query in natural language ('who do I know at NVIDIA?' or 'who haven't I talked to in a while?'). Auto-filter noise senders like marketing emails and newsletters. Build profiles for each contact with their company, role, how I know them, and our interaction history. Add relationship health scores that flag stale relationships, follow-up reminders I can create, snooze, or mark done, and duplicate contact detection with merge suggestions. Link relevant documents from Box to contacts so when I look up a person, I also see related docs."

2. Meeting Action Items (Fathom)

radiofun

#include <metal_stdlib>
#include <SwiftUI/SwiftUI.h>
using namespace metal;

[[ stitchable ]] half4 simpledistort(float2 pos, SwiftUI::Layer l, float4 boundingRect, float progress, float distortion) {
    
    float2 size = boundingRect.zw;
    float2 uv = pos / size;
    float2 center = float2(0.5, 0.5);
    float2 delta = uv - center;

karpathy

"""
The most atomic way to train and run inference for a GPT in pure, dependency-free Python.
This file is the complete algorithm.
Everything else is just efficiency.

@karpathy
"""

import os       # os.path.exists
import math     # math.log, math.exp

mberman84

# OpenClaw Implementation Prompts

Each prompt below is a self-contained brief you can hand to an AI coding assistant (or use as a project spec) to build that use case from scratch. Adapt the specific services to whatever you already use — the patterns are what matter.

---

## 1) Personal CRM Intelligence

```
Build me a personal CRM system that automatically tracks everyone I interact with, with smart filtering so it only adds real people — not newsletters, bots, or cold outreach.

davila7

# ============================================
  # Ghostty Terminal - Complete Configuration
  # ============================================
  # File: ~/.config/ghostty/config
  # Reload: Cmd+Shift+, (macOS)
  # View options: ghostty +show-config --default --docs

  # --- Typography ---
  font-family = JetBrainsMonoNerdFont
  font-size = 14

kieranklaassen

Claude Code TeammateTool - Source Code Analysis

This is not a proposal. This documents existing but hidden functionality found in Claude Code v2.1.19 binary, plus speculation on how it could be used.

---

Executive Summary

TeammateTool already exists in Claude Code. We extracted this from the compiled binary at ~/.local/share/claude/versions/2.1.19 using strings analysis. The feature is fully implemented but gated behind feature flags (I9() && qFB()).

cameroncooke

#!/usr/bin/env bash
set -euo pipefail

source_dirs=(
  "/Users/cameroncooke/Developer/dimillian-skills"
  "/Users/cameroncooke/Developer/sentry_skills"
  "/Users/cameroncooke/Developer/cameroncooke-skills"
  "/Users/cameroncooke/Developer/openai-skills/skills/.curated"
)