carloluisito

Stoke — Development skill for Claude Code

Development community

stoke is a local proxy that sits between Claude Code (or any Anthropic API client) and api.anthropic.com.

How to install Stoke

This entry records only its repository, not the path inside it, so there is no exact command to give. Open carloluisito/stoke and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Stoke does

stoke is a local proxy that sits between Claude Code (or any Anthropic API client) and api.anthropic.com. While you're idle — meetings, lunch, deep thought — it fires max_tokens: 0 keep-alive pings against your cached prefix just before the TTL expires. The cache stays warm. Your next prompt is cheap.

Alternatives in Development

Notion Meeting Intelligence — Prep meetings by pulling Notion context and tailoring agendas 14.6k ★
Claude Context Mode — An MCP server that sits between Claude Code and these outputs 5.3k ★
Om Meeting — Prep for any meeting by topic — subject-forward briefing with open tasks, blockers, and brainstormed considera 4.6k ★

README

stoke

One tool to minimize Claude Code token cost: keep the prompt cache warm, watch every dollar, and change wasteful behavior — with one command and one dashboard.

[![test](https://github.com/carloluisito/stoke/actions/workflows/test.yml/badge.svg)](https://github.com/carloluisito/stoke/actions/workflows/test.yml) [![license: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](./LICENSE)

**stoke** is two engines behind one CLI:

The proxy (packages/proxy, port 9876) sits between Claude Code and api.anthropic.com. While you're idle — meetings, lunch, deep thought — it fires max_tokens: 0 keep-alive pings against your cached prefix just before the TTL expires. The cache stays warm; your next prompt is cheap.
The monitor (packages/monitor, port 5599) parses every Claude Code transcript into SQLite, prices each turn, detects waste (cache expiries, context bloat, verbose output, model mismatch), and injects cost-saving hooks, skills, and a statusline into Claude Code itself.

They share one pricing source (`packages/shared`) and one database, so the dashboard shows one truthful number:

net cost = what Claude spent + what pings cost − rebuilds the proxy prevented

Why the proxy pays for itself

Anthropic's prompt cache has a TTL — **5 minutes** on API-key auth, **1 hour** on a Claude subscription. Walk away past that window and the next turn pays a full **cache rebuild at 1.25× the input rate** (2× on 1-hour TTL) — about $0.56 on a typical 150k-token session. stoke pays the **0.1× cache-read rate** (~$0.045) just before each expiry instead. The dashboard's headline is **net savings** (gross minus ping spend), never gross.

Install

git clone https://github.com/carloluisito/stoke
cd stoke
npm run setup

One command: installs dependencies, builds the dashboard, runs the full test suite as a gate, detects every Claude Code profile on the machine (`~/.claude`, `~/.claude-work`, `~/.claude-personal`, …),