Most project docs are written for humans. So an agent either rereads whole files every time it needs context, or skips the docs entirely and re-derives everything from source code. Neither scales well.
OKF fixes this by giving knowledge a shape an agent can navigate step by step. It gets an index.md it can skim before deciding what to open, type-tagged frontmatter it can filter on, and a log.md it can check for recent history, instead of one big wall of prose it has to read in full every time.
