claude-dashboard turns the Claude Code status line into an instrument panel

A status line that grew into a dashboard

claude-dashboard is a status line plugin for Claude Code, and the framing in the description is unified usage monitoring across the Claude, Codex, Gemini, Antigravity and z.ai command line tools, with context, rate limits, cost tracking and a modular widget system. The interesting part is the direction of travel. A status line is normally a single line of text, and this project starts there and keeps going. The requirement is Claude Code version 1.0.80 or later and Node.js 18 or newer. The architecture that makes it possible is the widget system, and the readme documents it as a table with categories and widgets, which reads as a specification of everything the plugin can show. The categories are core, rate limits, session, activity, analytics, multi CLI, insights, status and info, and the entries are specific enough to be useful as a reference for what is knowable about a coding session from the outside.

Three densities, and the default is one line

The display mode is a single setting with three values, and the differences between them are the most practical thing in the readme. The compact mode is one line and is the default, and it carries the model, a context progress bar, the cost, and either the rate limits for the five hour and seven day buckets or the z.ai usage, which are mutually exclusive by provider. The normal mode is two lines and adds the project information, the session identifier, the session duration, the burn rate and the todo progress. The detailed mode is six lines and adds the estimated depletion time, configuration counts, tool and agent status, the cache hit rate, a performance badge, a token breakdown, a forecast, the budget, and the usage of the other command line tools. The choice is essentially a question of how much of your attention you want to spend on the status line, and the fact that the compact mode is the default says the author expects people to want their prompt back most of the time.

The widget table is a catalogue of what a session exposes

Reading the widget table top to bottom is like reading an API surface. The core group starts with the model widget, which shows the model name and version with an emoji, the effort level for certain model families, and a fast mode marker, reading the live effort and fast mode values from standard input and falling back to the settings file on older versions of Claude Code. The context widget is a progress bar, a percentage and a token count, with the bar coloured by range, and it has sub-widgets for the bar alone, the percentage alone and the token count alone, so you can compose your own line. The cost widget shows the session cost in dollars, and project info shows the directory and git branch as a clickable terminal link, with ahead and behind counts, a subpath and a worktree indicator. Then the rate limit widgets, one for the five hour bucket with a reset countdown, one for the seven day bucket, and separate ones for the Sonnet and Fable variants, the latter of which is marked deprecated because the API no longer returns a separate field for it. The session group covers identifiers in short and full form, the session name, the duration and the last prompt with a timestamp.

Usage from four other command line tools

The multi CLI group is the part that makes this more than a status line. There are widgets for OpenAI Codex usage, for the Google Gemini CLI showing the current model or all models, for the Google Antigravity CLI showing the weekly quota by model family or per model, and for z.ai usage. The footnotes at the bottom of the table explain the behaviour in detail, and the behaviour is sensible rather than clever. Each widget hides itself when the corresponding tool is not present, detected by looking for the tool's credential file, such as the Codex authentication file for Codex and the Gemini credentials file for Gemini, and the z.ai widget only appears when the base URL environment variable points at that service. The Gemini footnote also records a change in the world, that the personal tiers of the Gemini CLI were retired in mid 2026 in favour of the Antigravity CLI, with enterprise still supported, which is why there is a family of Antigravity widgets and a note about the older one.

The prompt cache widget depends on a recent Claude Code release

One widget is worth singling out because it shows the plugin tracking a specific upstream behaviour. The prompt cache widget reports the health of the session cache, whether it is warm with the time left before it goes cold or cold, along with the hit ratio and a miss count, and the data comes from the prompt cache field that Claude Code exposes from version 2.1.251 onwards. It also explains that the warm to cold switch happens on its own, and that you can set the status line refresh interval, for example to sixty seconds, to keep the minute countdown ticking while you are idle. That last detail is the kind of thing that only matters to someone who has watched a countdown freeze, and it is also a hint about the plugin's architecture, since a status line that only updates on the next prompt cannot count down at all. The sub-widgets let you show only the icon and the time, only the hit ratio, or only the miss count, with the miss counter hidden when it is zero.

Marketplace install works; manual install has a catch

Installation has two routes and the readme is unusually clear about the difference. The marketplace route is three commands, adding the marketplace, installing the plugin and running the setup command.

/plugin marketplace add uppinote20/claude-dashboard
/plugin install claude-dashboard
/claude-dashboard:setup

The manual route clones the repository into the plugins directory of the Claude configuration, and then the note explains the catch in detail. The setup command looks for the plugin under a cached path, which a manual clone into the plugins directory does not match, so its shim lookup finds nothing to copy. The workaround is to point the status line command in the settings file directly at the built entry point in your clone. And the second catch is that manual installations do not get automatic version pickup, so you pull the changes yourself and update that path after each release.

git clone https://github.com/uppinote20/claude-dashboard.git "${CLAUDE_CONFIG_DIR:-$HOME/.claude}/plugins/claude-dashboard"

That is unusual candour for a readme, and it saves anyone who chose the manual route from a confusing afternoon.

A small build, a real test suite and a manual smoke test

The build is deliberately plain. The package is an ES module with the built entry point as its main file, the build script runs a Node script that bundles with esbuild, and the type checking is TypeScript. The test suite runs on vitest, and there is a watch mode alongside it. The most useful script for a contributor is the manual test, which pipes a JSON payload into the built binary and prints what the status line would render, so you can see the effect of a change to a widget without launching Claude Code at all. The payload includes a model display name, a context window size with the current usage broken into input, cache creation and cache read tokens, and a total cost, which is enough to exercise the core widgets.

The repository also carries a plugin manifest directory, a commands directory, hooks, a locales directory, a types directory, a website directory, a contributor guide and a changelog, which is a fairly complete project skeleton for something that is, in the end, one process that prints a line of text a few dozen times a minute.

Editorial conclusion

claude-dashboard earns its place for a heavy user of Claude Code who wants to see context, spend and rate limit state without leaving the terminal, and for a team running several coding agents at once, since the multi CLI widgets give one status line a view across all of them. It is the wrong choice if you want a minimal prompt, since the compact mode is already a dense single line and the detailed mode is six, and it is the wrong choice if you cannot run a plugin that reads the other tools' credential files. Verify three things before you install it: the Claude Code version you run, since the prompt cache widget needs a recent release and the plugin states a minimum version, whether you want the automatic update path from the plugin marketplace, because a manual clone does not pick up new versions on its own, and which of the multi CLI widgets you actually want, since some of them read credential files from other tools on your machine. The licence is MIT, the newest release is v1.35.1 from 2026-10-06, and the last push was on 2026-10-16.

Frequently asked questions

Official sources

1. License: MIT
2. Project website
3. README
4. Releases
5. uppinote20/claude-dashboard on GitHub