Understand Anything: Claude Code plugin for codebase graphs
A Claude Code plugin that maps your codebase into an interactive graph you can explore, search and ask questions about.
- Skills
- 9
- included
- Hooks
- 2
- included
- Agents
- 10
- included
- Stars
- 89.6K
- on GitHub
- Included
- 9
- listed
Understand Anything makes Claude Code read your project and build an interactive map of its files, functions and how they connect. You explore the map in a local dashboard, ask questions about the code, and get guided tours and onboarding guides. The first run reads the whole project, so it can use a lot of tokens on a large repo, and later runs only redo what changed. It is not in Anthropic's official marketplace, so install it from the author's repo with the command in the install box.
Good for
- Developers joining a project who need to see how the code fits together
- Teams that want an onboarding guide for new hires
- Reviewers who want to know what a change touches before they merge
- People working in a big or unfamiliar monorepo
- Anyone who can spend tokens up front for a reusable map
Install understand-anything
Claude Code
- 1
Copy this prompt.
Install the understand-anything plugin in Claude Code. Run this command: /plugin marketplace add Egonex-AI/Understand-Anything /plugin install understand-anything Then check that it loaded and tell me how to use it.
- 2
Paste it into Claude Code. It shows the change and asks before running it.
Requires Node.js 22 or later, pnpm 10 or later.
The first /understand builds the plugin and then reads your whole codebase, which can use a lot of your Claude usage on a large repo.
From the project README. Source.
Key facts
- License
- MIT
- Version
- 2.9.8
- Last commit
- October 26, 2026
- Created
- March 15, 2026
What it includes
Builds the interactive graph of your codebase
Reads your project, finds files, functions, classes and imports, and has the model write summaries, layers and tours. Saves the result as .ua/knowledge-graph.json. Later runs only redo changed files.
Try: “/understand src/frontend”
Runs code ·
/understandOpens the graph in a local web page
Starts a local web server on your machine and opens the graph in your browser. Use the address it prints, which includes an access token.
Try: “/understand-dashboard”
Runs code ·
/understand-dashboardAnswers questions using the graph
Looks up the graph to answer questions about how your code works.
Try: “/understand-chat How does the payment flow work?”
Files ·
/understand-chatShows what a change touches and the risks
Checks your uncommitted changes or a pull request against the graph and lists affected parts and risks.
Try: “/understand-diff”
Runs code ·
/understand-diffExplains one file, function or module
Gives a deeper walk-through of a single piece of code using what the graph knows around it.
Try: “/understand-explain src/auth/login.ts”
Files ·
/understand-explainWrites an onboarding guide
Produces a guide for a new teammate from the graph: layers, key parts and where to start.
Try: “/understand-onboard”
Write ·
/understand-onboardMaps business flows and process steps
Extracts business domains and flows from the code into a separate domain graph.
Try: “/understand-domain”
Write ·
/understand-domainTurns a notes wiki into a graph
Builds a graph of entities, links and topic clusters from a folder of LLM-maintained wiki notes.
Try: “/understand-knowledge ~/path/to/wiki”
Write ·
/understand-knowledgeBuilds a design graph from a Figma file
Reads a Figma file through the Figma API and builds a design graph. Needs a FIGMA_TOKEN and is the one skill that is not fully offline.
Try: “/understand-figma”
Write ·
/understand-figmaUsed by the assistant behind the scenes
Checks if the graph is out of date
A shell check at session start. If auto-update is on and the saved commit differs from git HEAD, it tells Claude to update the graph without asking. Otherwise it prints nothing.
Runs code ·
SessionStart hookRefreshes the graph after a commit
After each Bash call, a Node script checks for git commit, merge, cherry-pick or rebase. If auto-update is on and a graph exists, it tells Claude to update the graph.
Runs code ·
PostToolUse hook (Bash)Finds files, languages and frameworks
A helper agent that /understand and related skills call during analysis.
Read ·
project-scanner agentExtracts functions, classes and imports
A helper agent that /understand and related skills call during analysis.
Read ·
file-analyzer agentWorks out the architecture layers
A helper agent that /understand and related skills call during analysis.
Read ·
architecture-analyzer agentBuilds guided learning tours
A helper agent that /understand and related skills call during analysis.
Read ·
tour-builder agentChecks the graph is complete and valid
A helper agent that /understand and related skills call during analysis.
Read ·
graph-reviewer agentExtracts business flows and steps
A helper agent that /understand and related skills call during analysis.
Read ·
domain-analyzer agentReads wiki articles for entities and claims
A helper agent that /understand and related skills call during analysis.
Read ·
article-analyzer agentChecks merged batch graphs and recovers dropped parts
A helper agent that /understand and related skills call during analysis.
Read ·
assemble-reviewer agentAdds summaries and tags to a Figma design graph
A helper agent that /understand-figma calls to describe screens and components.
Read ·
design-analyzer agentHelps you query and read the graph
Explains graph structure, layers and tours when you ask how to use the graph.
Read ·
knowledge-graph-guide agent
From the official docs
Permissions
Medium risk
Its hooks and first-run commands can change your project and your session without much warning. Keep auto-update off unless you want that, and read each shell prompt on the first run.
What it can do
- Read your data
- Create and change things
- Delete things
- Read-only modeHow to turn it on
Watch out for
Two hooks run on every session and Bash call
The SessionStart hook is a shell check and the PostToolUse hook (on Bash) is a Node script. Both only read .ua/ files and git state, and with auto-update on they tell Claude to refresh the graph without asking. github.com
First run installs packages and asks a lot
It runs a package install and builds the plugin, and people report dozens of shell approval prompts. Read the prompts rather than approving everything blindly. github.com
The dashboard is a local web server
It listens on 127.0.0.1 and needs the token in the address it prints. Open that exact address and do not share it. github.com
The Figma skill sends your token to Figma
Only /understand-figma calls api.figma.com. Skip it if you do not want a Figma token in your environment. github.com
Auto-update is off until you turn it on
The hooks stay silent unless auto-update is set in .ua/config.json and a graph exists. Leave it off if you want to approve every refresh. github.com
Your source goes to your assistant's model
Claude (or your other assistant) reads your files to write the summaries. For private code, point it at a local model. github.com
The one-line installer clones and links files
For non-Claude clients, install.sh clones the repo to ~/.understand-anything/repo and links skills into the client's skills folder. Read the script before you pipe it to bash. github.com
Security advisories
No published security advisories as of Nov 2, 2026.
Technical details
| Turning on read-only | To only view an existing graph, run the standalone viewer from the latest release with the project path. It serves read-only from local disk and makes no model calls. Analysis itself always writes .ua/. |
|---|---|
| Talks to | api.figma.com (only /understand-figma, with your FIGMA_TOKEN). GitHub releases (dashboard viewer download via npx). Your assistant's model provider (it reads your source to write summaries) |
| Sign-in and keys | No account. The dashboard uses a token in the URL it prints. FIGMA_TOKEN is read from your environment only for the Figma skill. |
| Data retention | The graph stays in .ua/knowledge-graph.json in your project. SECURITY.md says the plugin does not phone home. |
| Scopes |
|
Works in
Claude Code is the native target. Other assistants install through the README's script or plugin commands.
Tap an app to see how to set it up and any known problems.
Cursor
SupportedFound in a cloned repo, or paste the repo URL in Settings > Plugins. GitHub
Copilot CLI
SupportedInstall with the copilot plugin install command from the README. GitHub
Gemini CLI
SupportedInstall with the one-line script, then restart. GitHub
Cline
SupportedInstall with the one-line script; skills link into ~/.cline/skills. GitHub
Claude Code
Has issuesNative plugin install from the author's marketplace.
Known problems
- Open
Claude often ignores the auto-update instruction at session start
Workaround: Run /understand after commits; it is incremental.
Since Jun 2026 · GitHub
- Open
Dashboard quick-start download fails, so it falls back to a slower local build
Since Sep 2026 · GitHub
- Open
First full analysis can use up a 5-hour or weekly limit on large projects
Since May 2026 · GitHub
- Open
Startup hook errors on Windows because it targets a Unix shell
Since May 2026 · GitHub
- Open
First run asks for shell approval dozens of times
Workaround: The maintainer suggests auto-accept for the run.
Since Apr 2026 · GitHub
Codex
Has issuesInstall with the one-line script, then type $understand.
Known problems
- Open
On Windows, Codex cannot find the agent files without a full path
Since May 2026 · GitHub
VS Code (Copilot)
Has issuesNeeds GitHub Copilot 1.108+; found in a cloned repo.
Known problems
- Open
Skills and agents are not detected after opening the cloned repo
Since May 2026 · GitHub
OpenCode
Has issuesInstall with the one-line script, then restart.
Known problems
- Open
Analysis expects Claude Code style subagents; OpenCode falls back to general agents
Since May 2026 · GitHub
What people say
People like the answers once the graph exists and dislike how many tokens and prompts the first run takes.
What people like
- Follow-up commands like chat and explain give richer answers than plain Claude Code, one user says.
- Better Stack found the architecture graph, guided tours and dependency path finder useful on a microservices demo.
- Free, MIT licensed, and later runs only redo changed files.
Common complaints
- The first run is slow and token-heavy, and users report hitting 5-hour and weekly limits on large repos.
- Dozens of shell approval prompts make the first run long.
- The graph is a snapshot that has to be rebuilt as the code changes.
- Some Hacker News commenters doubt a big node graph beats reading the code or asking the assistant directly.
In their words
“I'm using opus4.7 xhigh in Claude code and have only reached stage 2. My 5-hour quota has been exhausted. This consumption of tokens is too much.”Xu-ywGitHub issueApr 2026
“the subsequent commands like `/understand-chat` and `/understand-explain` return much richer and more useful results compared to using plain Claude Code.”Antoliny0919BlogJul 2026
“For the Online Boutique demo, this took approximately 30 minutes and consumed around 200,000 tokens.”Stanley UliliBlogJul 2026
“The token cost is the main practical constraint.”Stanley Ulili
Included in this plugin
9
- understand
Analyze a codebase to produce an interactive knowledge graph for understanding architecture, components, and relationships
5.5K installs - understand-chat
Use when you need to ask questions about a codebase or understand code using a knowledge graph
3.9K installs - understand-dashboard
Launch the interactive web dashboard to visualize a codebase's knowledge graph
4K installs - understand-diff
Use when you need to analyze git diffs or pull requests to understand what changed, affected components, and risks
3.8K installs - understand-domain
Extract business domain knowledge from a codebase and generate an interactive domain flow graph. Works standalone (lightweight scan) or derives from an existing /understand knowledge graph.
3.8K installs - understand-explain
Use when you need a deep-dive explanation of a specific file, function, or module in the codebase
4K installs - understand-figma
Analyze a Figma file via the Figma REST API and generate an interactive design knowledge graph (pages, screens, components, component sets, instances, design tokens) with a kind:"design" dashboard.
90K GitHub stars - understand-knowledge
Analyze a Karpathy-pattern LLM wiki knowledge base and generate an interactive knowledge graph with entity extraction, implicit relationships, and topic clustering.
3.6K installs - understand-onboard
Use when you need to generate an onboarding guide for new team members joining a project
3.9K installs
Questions
How do I install Understand Anything in Claude Code?
Use the install box above: add the author's marketplace, then install the plugin. Then run /understand in your project to build the graph and /understand-dashboard to open it. The first run needs Node.js 22+ and pnpm 10+.
How many tokens does it use?
It depends on project size. Better Stack reports about 30 minutes and around 200,000 tokens for Google's Online Boutique demo, and one user on a large repo ran out of a 5-hour Claude limit before the second stage. Later runs only redo changed files.
How do I cut the cost of the first run?
Point it at one folder, such as /understand src/frontend, and skip files with --exclude or a .understandignore file. A local model also works. One user in issue #212 had Claude run the sub-tasks on Sonnet and keep only the final step on Opus.
What is the dashboard access token?
The dashboard runs only on your machine and prints an address that ends in a ?token= value. You need that token to see the graph, so open the exact address it prints and do not share it.
Is Understand Anything safe?
It is MIT licensed and its SECURITY.md says it runs locally and does not phone home. Your code is still read by your assistant's model, two hooks run in Claude Code, and the first run installs packages. No security advisories are published for the repo.
Does it send my code anywhere?
Not to the author. Your files go to whichever model your assistant uses, so use a local model for private code. Only the Figma skill calls another service (Figma), and the dashboard may download a viewer from GitHub.
Does it work in Codex, Cursor or Gemini CLI?
Yes, according to the README. Codex, Gemini CLI, OpenCode and Cline install with the one-line script, Cursor and VS Code with Copilot find it in a cloned repo, and Copilot CLI has its own install command. In Codex you type $understand instead of /understand. It is not documented for the Claude app.
What happened to Lum1104/Understand-Anything?
The repo moved to egonex-ai/understand-anything, and the old address redirects there. Older posts, videos and issues still link to the Lum1104 name. It is the same project.
Where is the graph stored, and can I share it?
In .ua/knowledge-graph.json inside your project (older projects use .understand-anything/). It is plain JSON you can commit so teammates skip the build. Use git-lfs if it passes 10 MB.
Understand Anything vs graphify?
Both turn a codebase into a knowledge graph from a slash command. Understand Anything centres on a visual dashboard, guided tours and onboarding guides. graphify centres on a queryable graph that also takes docs, SQL schemas and PDFs. Try the smaller one on your own repo to compare the token cost.
Information
updated Nov 2, 2026
- Runs in
- Claude Code
- Repository
- egonex-ai/understand-anything
Sources
Docs, issue trackers and discussions this page is built from.
Show all 30Hide
- Plugin marketplace · github.com
- Understand-Anything README · github.com
- Copilot CLI plugin · github.com
- Cursor plugin · github.com
- Dashboard viewer only · github.com
- GitHub Copilot plugin · github.com
- Understand Anything: AI-Generated Knowledge Graphs for Large Codebases · betterstack.com
- claude-plugins-official marketplace.json · github.com
- plugin.json · github.com
- understand SKILL.md · github.com
- Plugin folder · github.com
- understand-dashboard SKILL.md · github.com
- understand-figma SKILL.md · github.com
- Plugin hooks.json · github.com
- Plugin agents folder · github.com
- post-tool-use-auto-update.mjs · github.com
- Understand-Anything SECURITY.md · github.com
- install.sh · github.com
- How to make /understand run faster? · github.com
- token consumption problem · github.com
- Egonex-AI/Understand-Anything on GitHub · github.com
- graphify on GitHub · github.com
- token consumption problem (comment) · github.com
- Understand Anything (Hacker News) · news.ycombinator.com
- agents ignore hooks instructions · github.com
- Dashboard fast path 404s for every plugin version after 2.9.0 · github.com
- Windows incompatibility · github.com
- [Windows] Codex fails to locate agents unless full path is specified · github.com
- VSCode: Copilot plugin not detected · github.com
- OpenCode issues with subagent-based analysis phase · github.com