By name: understand-codebase description: "Use when: onboard onto an unfamiliar or large codebase by mapping its structure, dependencies, and business logic before editing." license: MIT
Goal: turn a codebase into a navigable knowledge graph you can explore, search, and explain — not a graph that just looks complex.
Use for:
- joining a large unfamiliar project and needing to know where to start
- learning how files, functions, and classes connect before changing one
- mapping code to business domains, flows, and process steps
- assessing which parts a change ripples into before committing
Workflow:
- Scan the project: detect languages and frameworks, discover every file.
- Extract structure deterministically (tree-sitter): imports, exports, function/class definitions, call sites, inheritance. Same input -> same edges every run.
- Layer in semantics (LLM): plain-English summaries, architectural layer per node (API / Service / Data / UI / Utility), business domains and flows.
- Build the graph — nodes are files/functions/classes, edges are dependencies.
- Generate guided tours ordered by dependency so the codebase is learned in the right order.
- Re-run incrementally: only re-analyze files whose fingerprint changed.
Outputs:
- structural + domain graph with nodes searchable by name and by meaning
- guided tours, layer view, and an onboarding guide for new contributors
- diff impact: which parts a current change affects
Rules:
- Keep structure deterministic; reserve the LLM for intent, summaries, and layering.
- Scope to a subdirectory for huge monorepos.
- Commit the graph so teammates skip the pipeline; refresh it on changed files.