Understand Anything is a Claude Code plugin that analyzes a project and builds a knowledge graph of its files, functions, classes, and dependencies, then opens an interactive dashboard to explore it. It is aimed at developers joining an unfamiliar codebase: selecting a node shows its code, relationships, and a plain-English explanation, and guided tours walk through the architecture in dependency order. The same skills also install into Codex, Cursor, GitHub Copilot, Gemini CLI, OpenCode, OpenClaw, Pi, Hermes, Cline, Kimi CLI, Nanobot, Kiro, and other coding platforms.
Analysis combines Tree-sitter and an LLM. Tree-sitter parses source deterministically to extract imports, exports, definitions, call sites, and inheritance, so the structural edges are the same on every run. The LLM reads that structure with the source to write summaries, tags, architectural layer assignments, business-domain mappings, and tours. A multi-agent pipeline (project scanner, file analyzer, architecture analyzer, tour builder, and graph reviewer, plus domain and article analyzers) runs file analysis in parallel, and later runs re-analyze only changed files.
Beyond code, a domain view maps code to business domains, flows, and process steps, and a knowledge mode turns a Karpathy-pattern LLM wiki into a clustered graph of entities, claims, and implicit relationships. The graph is plain JSON, so a team can commit it and teammates open the dashboard with a standalone viewer that serves it read-only from local disk, with no LLM calls. The initial analysis of a large project can use many tokens, and it can run on a local model provider such as Ollama.
Features
- Structural graph: every file, function, and class is a clickable node with summaries and relationships
- Business domain view: maps code to domains, flows, and process steps laid out as a horizontal graph
- Guided tours: generated walkthroughs of the architecture, ordered by dependency, plus an onboarding guide for new team members
- Fuzzy and semantic search: finds nodes by name or by meaning, such as which parts handle auth
- Diff impact analysis: shows which parts of the system the current changes affect before you commit
- Codebase chat: answers questions about the codebase and explains a specific file or function in depth
- Layer visualization: groups nodes by architectural layer (API, Service, Data, UI, Utility) with a color-coded legend
- Knowledge base graphs: analyzes a Karpathy-pattern LLM wiki into a force-directed graph with community clustering
- Beyond source code: includes Dockerfiles, Terraform, SQL, Markdown, and other file types in one graph
- Incremental updates: re-analyzes only changed files, optionally on every commit through a post-commit hook
- Shareable graph: the committed JSON opens in a standalone read-only viewer, and exports to PNG, SVG, or filtered JSON
- Localized output: generates summaries, dashboard labels, and tours in English, Chinese, Japanese, Korean, Russian, or Vietnamese
