The Knowledge Graph

# The Knowledge Graph

Ninety-one SKILL.md files on this machine. I could not tell you what half of them do.

That is not a filing problem. It is the reason I kept building things I already owned — a session
searcher, a repo summariser, a skills index, each built twice because the first one had gone quiet in
a folder I stopped opening.

GRAPH is the map that ended it.

## What it actually does

Reads the whole tree. Not a file listing — it opens the files.

Names what each thing actually is. This is the part that matters. A folder called utils/ tells
you nothing; the graph reads what's inside and labels it by behaviour, so the name on the map is the
thing rather than the directory someone happened to choose two years ago.

Wires the ones that touch. Relationships become queryable instead of remembered. That's the
difference between a list and a map.

"Which of these already solves this?" Asked before you build, this is the whole return on the
thing. Most duplicate work is not laziness; it's a memory failure about your own repo, and it starts
somewhere around file forty.

"What breaks if I delete it?" Asked before you clean up. The reason dead code survives for years
is that nobody can cheaply prove it's dead.

Both answered from the map instead of from memory.

Every other kit I've published runs on the standard library and costs nothing. This one doesn't.

The graph calls a paid model on every build, and it refuses to run without your own API key
(OPENROUTER_API_KEY). It fails closed rather than silently degrading, which is the right behaviour
— but it means this is a tool with a running cost, and you should know that before you unzip it and
not after.

That property is also why a spend gate belongs in the harness rather than in your intentions: a
paid-API pipeline with no declared cap is a bill waiting for a bad night.

- SKILL.md — build, query, path, explain: the four things you actually do with it
- LEARNINGS.md — what went wrong in practice, with the incidents attached
- references/ — 9 docs covering the extraction pipeline, the graph model, and the query language
- README.md — install steps, the key requirement stated on the first screen, and which soft paths
to repoint for your own setup

The map is a snapshot: it is true about the tree at the moment you built it, so re-run it when the
tree changes. The README says that too.

Comment GRAPH on the video and I'll send it over — though nothing is gated on that. It's at
sharpendaily.co/kit/graph either way, no email and no DM.