Dashboards·terminal · desktop·MIT
netrunner-hud
Animated cyberpunk HUD: context gauge, token oscilloscope, tool waterfall, git and cost.

netrunner-hud opens a cyberpunk dashboard for your Claude Code session in its own pane, animated at about 30 frames a second. At a glance you see how full the context window is, how fast Claude is writing, which tools it called and how long each one took, plus git, test, background job and cost readouts. The /hud command adds nothing to the conversation, so the dashboard costs no tokens.
What it does
- Header. A session clock labelled UPLINK, context fill, session cost and the current token rate, with a magenta-to-cyan sweep, a moving scanline and a brief katakana glitch every few seconds.
- Context gauge. A braille semicircle that fills from green through amber to red as the context window fills, with a tick every 10% and a pulsing needle. Its label turns red past 80%.
- Token flux. A scrolling braille waveform of Claude’s output rate, counting text, thinking and tool input at roughly four characters per token, sampled ten times a second. Older trace fades out, and the peak sits beside the current rate. It only counts while the HUD is open.
- Tool waterfall. One row per tool call: when it started, a bar whose length is the duration on a log scale (a minute fills it), the time in seconds and a short label such as the Bash command, the file name, the search pattern or a URL’s host. Bash is green, edits amber, reads and searches cyan, web and MCP tools violet, subagents magenta. Failed calls turn red with
ERR, and calls made by a subagent get a↳. - Widgets. Background jobs Claude started, git branch and number of changed files, the last test run, and session cost with an hourly burn rate.

The git widget shows status unknown when git is busy or fails, never a false clean. The tests widget spots the usual runners when they are the program a command actually runs, such as npm test, pytest, go test, cargo test, make test, jest, vitest or playwright test, and ignores list, install and help modes. It shows PASS or FAIL for a plain test run; after a leading cd dir && only a pass is certain, and a runner buried in a longer compound command shows ?. A background job leaves the widget when its completion notice arrives or Claude stops it.
In a wide terminal the pane docks beside the transcript; in a narrow one it opens inline. The waterfall shows as many recent calls as fit, up to 20.
Install
Needs Claude Code 2.1.287 or later, plus a terminal pane at least 60 columns wide for the animated view.
claude plugin marketplace add ccdwyer/claude-mods
claude plugin install netrunner-hud@ccdwyer-mods
The marketplace is the author’s shared one, which also carries codebase-galaxy and context-dungeon; if you already added it, the second command is enough.
Run /reload-plugins, then /hud to open the pane. Run /hud again to close it; the animation only runs while the pane is open.
How it works
The tool.call hook notes each call’s tool, its short label, how long it ran and whether it failed, then passes the result on unchanged. For Bash calls it also checks the command for a test runner or a git command and keeps the ID of any background task Claude starts. The turn.step hook peeks at Claude’s streamed reply only to count characters for the oscilloscope, and only while the HUD is open; every chunk goes through untouched. The prompt.submit hook looks only at background task notifications, to clear the finished job.
While the pane is open, it reads the session’s usage figures (context fill and cost) once a second and runs one local git status in the session’s folder every five seconds, plus once after any Bash command that calls git. That git status turns off the repository’s status alias and fsmonitor hook, takes no optional locks and gives up after 2.5 seconds. Usage and git run on these slow timers, separate from the frames, so a slow repository never stalls the animation.
The last 60 calls, up to 16 background jobs and the latest readings stay in the plugin’s session state. It writes no files and makes no network calls, which matches the README’s privacy section.
Limitations
- The animated view needs a terminal. In Claude Desktop’s Code tab, or in a pane narrower than 60 columns, the HUD shows a compact text summary instead: context, cost, git, tests, job count and the last eight tool calls. The animation stops there, but the readings keep refreshing.
- The token rate is an estimate from character counts, not a tokenizer, so treat the waveform as a pulse rather than a meter.
Similar Claude Mods
transit-map
Git history as a Vignelli-style subway map; the train moves when Claude commits.
Flightdeck
Subagents, permission checks, advisor consults, context and cost, live in a pane.
token-weather-usage
Context, 5-hour and 7-day limits, cache countdown and cost on one line above the prompt.
claude-carbon
Estimated grams of CO2 for each answer and session, with a molecule and warming stripes.
Popular Claude Mods
The whole directory →auto-handoff
Hands a long session to a fresh one with a Haiku-written brief instead of compacting.
ripple
Effort per prompt, subagents routed to GPT, DeepSeek or Claude, and cache warnings.
replay-theater
A /replay pane that walks through the last turn's file edits, one diff per step.
show-your-work
Claude shows screenshots and recorded Playwright runs in the chat, with before/after QA.







