Overview

xiaoyuze88/karpathy-llm-wiki is a zero-code, pure prompt-engineering boilerplate using a “Hot.md” focus cache. It is a Tier 4 starter template in the comprehensive report — the lightest possible setup.

Key Facts

  • Approach: zero-code, pure prompt engineering
  • Trick: “Hot.md” focus cache to keep context warm
  • Role: opinionated minimal boilerplate
  • Stack: Claude Code + prompts + Hot.md

Features

Core Architecture

  • Zero-code — pure prompt engineering, no scripts or tools
  • Hot.md focus cache — keeps frequently accessed context warm in LLM memory
  • Prompt-only — everything driven by CLAUDE.md instructions

Ingest & Processing

  • Agent-driven ingest — Claude Code processes sources via prompts
  • Focus cache — Hot.md maintains context across sessions

Integration

  • Claude Code native — designed for Claude Code prompt system
  • No dependencies — zero packages, zero databases

Data & Storage

  • Markdown files — standard wiki output
  • Hot.md — focus cache for context management

Relationships

  • llm-wiki — (type: belongs-to) prompt-only implementation
  • claude-code — (type: operates-with) driven entirely by prompts/CLAUDE.md
  • ronancodes-llm-wiki — (type: sibling-of) sibling boilerplate (RonanCodes is CLI-structured; this is prompt-only)

Implications

Shows that the pattern can run with no code at all — just prompts + a focus-cache file. Relevant to the user’s “too manual” concern only partially: it removes code, but the human still drives prompts.

  • Repository URL unavailable in the current freshness check; the implementation summary is preserved in raw/articles/llm-wiki-okf-comprehensive-report-2026.md.

Sources