Cache Tax

MIT licence

A Claude Code mod that keeps the one-hour prompt cache warm during breaks and stops a cold send once with its estimated rewrite cost. Warming uses model requests and costs tokens. Requires Claude Code 2.1.287+.

0 copies

Preview

Terminal

❯ /keepwarm 
  ⎿  cache-tax: keepwarm on for 6h00m, a ping 50m after each idle stretch keeps the cache read, not
     re-written
[>] cache-tax · warm · keepwarm 6h00m left · ping in 50m                                        [-]
❯

Claude Desktop

What it does

A row above the prompt that keeps the one-hour prompt cache warm during breaks, and stops a cold send once to show what rewriting the context will cost.

What you see: A status row above the prompt while keepwarm is on
Works in: Terminal and Claude Desktop
Slash commands: /cache-tax, /keepwarm
Uses your tokens: A one-line ping after 50 idle minutes in a window
Changes files: Nothing
Network access: None
Reviewed commit: 5cde9ac
Preview recorded with: Claude Code 2.1.296

How it works

/keepwarm arms a six-hour window, or a shorter one such as /keepwarm 90m
After 50 idle minutes in a window, it sends one short model request to keep the cache
Shows warm, cold or unknown above the prompt, with time left and the next ping
On a cold cache with 50,000+ tokens of context, it drops your send once with its price

Good to know

Every ping costs tokens, so pick a window you expect to come back within.
It needs a one-hour cache. On API-key or cloud billing, set promptCacheTtl to 1h first.
Send the dropped message again to pay the rewrite, or run /cache-tax guard warn to only see the price.

More like this

Token Weather

Mod

0 copies

TokensMinimal

Terminal

 ☀  Clear  22% of context  44k / 200k   last turns ██  steady                                   [-]
❯

Claude Desktop

A live forecast of the context window, drawn above the prompt: a weather icon, the fill percentage, and a chart of the last turns.

@anthropics

claude-usage

Status line

3 copies

LimitsMinimalPythonNetwork AccessReads Token

✳ Usage 5h 26% · week 7% · fable 15%

Your official Claude quota: the 5-hour session, weekly, and per-model windows with reset times, read from the same endpoint as /usage rather than estimated from token logs.

@Tatendaz

Context Burn Bar

Status line

0 copies

TokensMinimalBash

5h ▓░░░░ 🟢

A minimal 5-slot block bar showing context-window usage with a green/yellow/red status dot.

@gwiener