LLM knowledge base skills
A set of agent skills that turn a directory of markdown notes into a self-updating knowledge base of tagged, backlinked notes and generated wikis, following Andrej Karpathy's llm-wiki pattern.
These skills let an AI coding agent maintain a personal knowledge base without a human editing it directly. A user writes raw notes into a folder, and the enrich-note skill tags a note with topics from a shared registry, adds source attribution and links it to related notes, while enrich-notes-loop runs that skill unattended over every note that has not yet been enriched, tracked with a frontmatter timestamp. A refresh-wiki skill then maintains generated wiki pages under a wikis folder, ingesting new notes, updating entity and concept pages, and linting for stale claims or orphan pages.
Cloud-wrapped variants of the enrich and refresh skills sync an Obsidian vault down, run the same workflow, and sync it back up, and a setup skill walks the agent through creating the scheduled cloud environment. Two self-contained HTML files, dropped beside the notes, visualize the vault as an activity graph or a night-sky map clustered by tag.