AI Assistants
PhysioTwin4D is developed with AI coding assistants in the loop (Claude Code, Codex, Cursor, and others). Two repository conventions make that work well.
Repository Instructions
AGENTS.md holds the role-based guidance that applies across every
assistant; CLAUDE.md adds Claude-specific instructions. Role subagents live
in .agents/agents/ and slash-command skills in .agents/skills/. Read
those before hand-writing a prompt that repeats project conventions.
One of those conventions binds every assistant: PhysioTwin4D prefers compatibility, breaks a public API only when the change is generally beneficial to future users, and never ships deprecation shims. Any commit that does break an API must add an entry to Migration Guide — Unreleased describing the change and the code that automates the conversion.
graphify
The recommended way to explore this codebase with an AI assistant is the
graphify knowledge graph in graphify-out/, which indexes god nodes,
community structure, and cross-file relationships.
A graph query returns a scoped subgraph. That is dramatically smaller and more
accurate than pasting whole files or raw grep output into a prompt, which
matters here: the source tree spans segmentation, registration, statistical
modeling, and USD export, and the relevant call chain for a question is rarely
in one file.
graphify query "how does a labelmap become a USD stage?"
graphify path "WorkflowConvertImageToVTK" "ConvertVTKToUSD"
graphify explain "anatomy taxonomy"
Recommended use:
Ask
graphify querybefore searching manually, whenevergraphify-out/graph.jsonexists.Use
graphify pathto see how two classes relate, andgraphify explainto focus on a single concept.Browse
graphify-out/wiki/index.mdfor broad navigation; readgraphify-out/GRAPH_REPORT.mdonly for whole-architecture review.Run
graphify update .after changing code. The update is AST-only, so it costs no API calls, and it keeps assistants from resolving symbols against a stale graph.