Understand Anything Turns a Codebase into a Knowledge Graph You Can Query

You just joined a team whose codebase is 200,000 lines. Where do you even start? Understand Anything, an 83k-star Claude Code plugin, answers with a map instead of a reading list.

Why This Skill Matters

Run /understand and a multi-agent pipeline scans the project, then writes a knowledge graph — every file, function, class, and dependency — to .ua/knowledge-graph.json. The dashboard turns it into something you can pan, zoom, and search: nodes carry plain-English summaries, layers are color-coded (API, Service, Data, UI, Utility), guided tours walk the architecture in dependency order, and semantic search answers questions like "which parts handle auth?".

The pipeline splits the work by what each tool does best. Tree-sitter parses source deterministically — imports, exports, definitions, call sites — so the structural side is reproducible; LLM agents add what parsers cannot: summaries, tags, architectural layer assignments, and business-domain mapping. Five agents run /understand: project-scanner, file-analyzer, architecture-analyzer, tour-builder, and graph-reviewer. /understand-domain adds a domain-analyzer; /understand-knowledge adds an article-analyzer for wiki knowledge bases.

Installation

/plugin marketplace add Egonex-AI/Understand-Anything
/plugin install understand-anything

Other platforms take a one-line curl installer — Codex, Gemini CLI, OpenCode, Cline, and more. Note that Codex invokes the skills with $understand instead of a slash. Cursor and VS Code with Copilot auto-discover the plugin when the repo is cloned.

Real Workflow: Map a Codebase, Then Explore It

/understand

The first run analyzes the whole project — the README warns it can consume a significant number of tokens on large codebases — while later runs are incremental, re-analyzing only changed files. Then open the view:

/understand-dashboard

An interactive dashboard opens with the codebase as a searchable, clickable graph, color-coded by architectural layer. Add --language zh to generate node summaries, dashboard labels, and tour text in Chinese — en, zh, zh-TW, ja, ko, and ru are supported. On a first run without the flag, the plugin detects your conversation language and, if it isn't English, asks to confirm before generating; English conversations are unaffected. The choice is saved to .ua/config.json for every later run.

Real Workflow: Onboard a Teammate and Stay Current

/understand-chat How does the payment flow work?
/understand-diff
/understand-explain src/auth/login.ts
/understand-onboard

/understand-chat asks the graph questions in plain language. /understand-diff shows which parts of the system your pending changes touch before you commit. /understand-explain deep-dives one file or function; /understand-onboard generates a guide for the next hire. For teams, commit the .ua/ directory — everything except intermediate/ and diff-overlay.json — and any teammate can open the dashboard with the standalone viewer: no Claude Code, no LLM, just Node.js 18+.

Tips

• Huge monorepo? Scope the scan: /understand src/frontend.
• Enable /understand --auto-update to install a post-commit hook that patches the graph with each commit, or re-run /understand manually before releases.
• Graphs over 10 MB should move to git-lfs; the README shows the three commands.
• /understand-knowledge is not just for code — point it at a Karpathy-pattern LLM wiki to get a force-directed knowledge graph with community clustering.

When Not to Use This

This is static mapping plus semantic summary, not runtime observability: it shows the code as written, not latency, errors, or traffic. And the first full scan of a big repository is genuinely token-hungry — if the budget is tight, start with a subdirectory, or follow the README's suggestion to point your platform at a local model provider such as Ollama.

See the leaderboard for more skills.