A skill is a projection of a graph

From the 4 June architecture brief:

“a skill is a projection of a semantic knowledge graph, where today’s skills are static photographs of what they should be.”

Typed primitives

The decomposition: a skill is “a graph of graphs of graphs” of typed primitives — principles, concepts, facts, knowledge, actions — connected across multiple graph types (“not only semantic knowledge graphs but control-flow graphs, permission maps, authentication maps, identity graphs”), with the projection “created in the context of use (multilingual, cultural, situational).”

A SKILL.md file is one flattened rendering of that graph, fixed at write time, for one context. The theory’s complaint is exactly that fixing: the same underlying knowledge should project differently for a different language, a different regulatory context, a different caller — and today it cannot, because there is no graph underneath the markdown, only the markdown.

The forking ecosystem

“Forking a skill is good; sync back via pull-request-equivalents; edge customisation stays compatible with the top.”

This is the mechanism the economy depends on for safety: a corporate skill forks into team- or role-specific versions, and because the fork relationship is recorded rather than the copy being silent, drift is visible and a sync-back is possible. Compare that to what actually happened with use-sg-playwright: a silent copy, no recorded relationship, and a drift nobody could see until someone measured it by hand.

What this is not

“Meaning through connectivity” — the general theory behind treating knowledge as a graph rather than a document — is graphs.sgit.ai’s founding concept. This site owns the application of that idea to skills specifically; that site owns the idea itself.