/notebook-graph

description: Turn a notebook knowledge base into a knowledge GRAPH — a node-link graph.json + a self-contained interactive graph.html — with NO agy calls and NO extra deps. Derives edges from entity co-occurrence within documents (works even though agy rarely emits explicit relations) plus any explicit relations. See who/what connects to what, and the hub entities. Complements /agy:notebook-query (exact SQL) and /agy:graph (Graphify, for code).
argument-hint: "<folder>"
context: fork
allowed-tools: Bash, Read

Build a graph view of the `notebook.db` that `/agy:notebook` compiled — instantly, locally, free. agy rarely fills the explicit `relaciones` array, so this derives edges from **entity co-occurrence within each document** (the relational signal that needs no LLM) plus any explicit relations present.

Raw user request: $ARGUMENTS

Phase 0 — Build the graph (ONE Bash call)

Resolve `OUTDIR = docs/agy/notebook/<slug>` from the folder (same `slug()` the notebook uses). If `notebook.db` is missing, tell the user to run `/agy:notebook <folder> | <objetivo>` first. Then:

python "${CLAUDE_PLUGIN_ROOT:-$PWD/plugins/antigravity}/scripts/notebook_graph.py" "$OUTDIR"

It prints `GRAPH nodes=N edges=M (cooccur=…, explicit=…) hubs=[…]` and writes `<OUTDIR>/graph.json` (node-link) + `<OUTDIR>/graph.html` (self-contained interactive viz, vis-network from CDN).

Phase 1 — Present + reason

• Report the node/edge counts and the **hub entities** (most connected — your central people / orgs /

references). Point the user to `graph.html` for the interactive view.

• Nodes are deduped entities (`persona | organizacion | referencia`) + document nodes; edges are

`appears_in` (entity→document), `co_mentioned` (entities sharing a document, weighted by # docs), and any explicit relations.

• To answer a question, read `graph.json` directly (small) — e.g. neighbours of an entity, shortest

path between two, or the densest cluster — instead of re-reading the source documents.

Notes

• **Zero agy calls, zero new deps** — pure stdlib over the existing `notebook.db`. Re-run after a

fresh `/agy:notebook` to refresh.

• **vs `/agy:graph`**: this graphs the *document/entity* network from your notebook RAG (free, local);

`/agy:graph` uses Graphify (Gemini via agy) for richer graphs incl. **code** (tree-sitter ASTs) and Leiden communities. Use this for a quick document graph, `/agy:graph` for code or deep graphs.

• For exact aggregates (amounts, timelines) use `/agy:notebook-query`.