The Aion ecosystem looks at itself — a living 3D graph of every node, every dependency, every anomaly.
Atlas parses disk and Notion in parallel — every CLAUDE.md, every skill, every tool, every project becomes a node. Every cross-reference, cascade relation, Notion-disk mirror becomes a directed edge. The result: a JSON graph rendered in 3D via 3d-force-graph and Three.js, where the structural health of the ecosystem is visible at a glance.
The trunk is ~/.claude/CLAUDE.md. Orphan nodes drift to the periphery, cycles break gravity, broken pointers glow red. This is not a documentation tool — it is a reflexivity tool: Aion sees how it works, detects its own drifts, proposes its own fixes.
The FastAPI backend (port 8113) exposes a /api/refresh that re-runs the parser on demand. The frontend is served via the Mockup tool — lightweight window, no framework. The V2 (live CC orbits via Hermes) and V3 (Argus Traveler, a VLM that reads graph.md) versions are already stubbed in the architecture.
Node with no inbound edge (excluding root) — skill or tool created but never referenced in the ecosystem.
DFS detects dependency loops — when A loads B which loads A, the doctrine bites its own tail.
Reference in a .md to a missing file — dead documentation, path renamed without an update.
Disk folder with no matching Notion page, or vice versa — a break in the source-of-truth mirror.
Regex on known-dead phrasings — flags CLAUDE.md files that didn't follow the ecosystem's evolution.
Three.js + 3d-force-graph CDN — hover shows the name, click isolates the node and highlights its edges.