Notemd knowledge graph
Agent Skills that teach Claude and other AI agents the Markdown note.md actually renders, and how its knowledge graph is derived.
npx -y skills add a-aigner/notemd-skills --skill notemd-knowledge-graphAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- 15 days oldThe repository was created 15 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Use when the user asks about notemd's graph view, knowledge graph, backlinks, connections between notes, or "argument maps", or when reasoning about how notemd notes relate. Explains that notemd's graph is derived (not an authorable file) and that argument maps are read-only AI output that must not be fabricated.
SKILL.md
2.7 KB, 568 tokens by cl100k_base, as published. Nobody here has run it
notemd Knowledge Graph & Argument Maps (read-only reference)
notemd has two graph-like features. Neither is a file you author. This skill exists so an agent understands what they are, how to influence them, and — critically — what not to fabricate. There is no .canvas / JSON Canvas equivalent in notemd; do not create one.
The knowledge graph is derived, not authored
notemd's graph view is generated from the notes themselves. You cannot write nodes or edges directly. Edges come from three sources:
| Edge kind | Comes from |
|---|---|
articleLink | Wikilinks [[Name]] between notes (see notemd-markdown) |
hierarchy | Folder-note layout — Foo.md beside a Foo/ folder parents its contents |
citation | Citation links to indexed sources (notemd://cite/…, created in-app) |
To shape the graph, change the notes: add or remove [[wikilinks]], and organize folder notes. That is the only lever. Do not attempt to emit graph data structures.
Argument maps are read-only AI output
An argument map is not authored by hand or by an agent. notemd generates it in-app by running an extraction model over an indexed source document (typically a PDF). Its structure (ArgumentMapDocument):
- Nodes with kinds:
central_claim,sub_claim,evidence,assumption,counterargument,rebuttal,limitation,scope_condition,cited_warrant. - Edges with kinds:
supports,qualifies,rebuts,assumes,evidences. - Each node carries a verbatim quote anchor tied to a specific page of the indexed source, plus a confidence tier (
exact/supported/weak/none) derived from a post-hoc quote validator — not a self-reported score.
Never fabricate or hand-edit an argument map. Its trustworthiness rests entirely on those anchors being real quotes from indexed chunks; a hand-written map produces a confidence signal that is a lie. If the user asks to "make an argument map," direct them to notemd's in-app extraction over the source document.
How you CAN help
- Interpret / summarize an existing argument map or graph the user shows you.
- Improve connectivity by adding accurate
[[wikilinks]]between related notes and organizing folder notes — this enriches the derived graph legitimately. - Explain why a note is isolated in the graph (usually: nothing links to it and it links to nothing).