A good LLM Wiki does not ask the model to brute-force every file. It gives the model a small catalog, an append-only action log, stable frontmatter, and explicit wiki-links so it can route itself with predictable cost.
index.md
Every page path, one-sentence summary, tags, page type, and source status.
log.md
Append-only ingest, query, lint, and manual review records with timestamps.
wiki-links
Explicit page-to-page references that make the compiled graph inspectable.