llm-wikilisted

# LLM Wiki — Skill A self-maintaining knowledge base where the LLM, not a vector DB, is the index. Inspired by Andrej Karpathy's note-taking method. Works standalone in any repo that has the wiki layout below; pairs with the Telegram ingest bot (`scripts/tg_bot.py`) for a hands-free pipeline. ## Wiki layout this skill expects ``` raw/ # source notes — READ ONLY, never edit or delete _inbox/ # incoming forwarded posts (bot drops them here) _assets/ # images extracted from processed posts knowledge/ concepts/ # ideas, methods, techniques tools/ # utilities, libraries, plugins, services, CLIs, platforms (NOT skills) skills/ # Claude Code Skills only (SKILL.md packages in ~/.claude/skills/) connections/ # links between topics (X vs Y, X + Y, patterns) daily/ # auto session logs index.md # navigation index — updated on every INGEST log.md # append-only operation history (newest on top) ``` If `CLAUDE.md` exists at repo root, read it first — it holds the authoritative schema and language preference. ## Naming & format (non-negotiable) - File names: **kebab-case**, `.md`, Latin only, no spaces (`claude-code-hooks.md`). - Every `knowledge/` page starts with YAML frontmatter: ```yaml --- title: "..." tags: [..] created: YYYY-MM-DD updated: YYYY-MM-DD sources: [] # raw/ paths or URLs related: [] # other knowledge/ pages ---