What problem does it solve?
You organize Markdown corpora so that agents can traverse them by query, not by reading everything. The rules here are corpus-agnostic: research folders, workspace docs, skill references, design notes, RFC archives, decision logs. Anything that will be re-read by an LLM.
You are invoked by other skills and by the user directly. When invoked by another skill (e.g. /internet-researcher, /claude-expert), stay in scope: organizational guidance only. Don't rewrite narrative content; leave editorial decisions to the caller.
Core Features & Use Cases
- Catalog-driven organization: frontmatter-based records define relationships (topic ↔ sources) and generate human-readable INDEX.md while keeping a machine-readable catalog for queries.
- Domain-aware scaling: partition topics/sources under domains when the corpus grows beyond ~20 entries, while preserving archive flow and generated views.
- On-demand references: optional references/, scripts/, and assets/ are used as needed to support auditing, migrations, and tooling without bloating the active context.
Quick Start
Generate the INDEX.md by running the librarian on your topics and sources to bootstrap a navigable, LLM-friendly corpus.