Discover gists

Claude Code burns ~15–20% of your tokens on a 3-word status line. Here's how to turn it off

by Flamerix AI · MIT License

Tested on Linux, Claude Code 2.1.257–2.1.312 (see Tested versions). The binary is not modified (BUN_OPTIONS=--preload). An update can silently undo it, so see Verify.

TL;DR

curl -fsSL https://gist.githubusercontent.com/flamerix-ai/c3905705cc4205e5de3011fa2e74efce/raw/no-agent-summary.js -o ~/.local/share/no-agent-summary.js

Running Claude Code Locally with Ollama and Open-Source Models as a Free Alternative to the Anthropic API

Claude Code's API costs add up fast for heavy users, often $50 to $200+/month on Opus 4.5/4.6. Ollama (v0.14.0+) now supports the Anthropic Messages API natively, which means Claude Code can run against local open-source models at zero cost, with no data leaving the machine.

This guide covers the full setup: installing Ollama and Claude Code, choosing a model that fits 16 GB of RAM, connecting the pieces, and understanding the real tradeoffs.

The Cost Problem

name	explain-diff-html
description	Use when the user asks for a rich explanation of a code change, diff, branch, or PR. Produces HTML output.

Explain Diff

Please make me a rich, interactive explanation of the specified code change.

It should have these sections:

适用于编程 agent 的系统提示词，亲测能让各类（DeepSeek/Mimo/GLM/Qwen/Gemini）flash 模型表现得极其专业、克制、敏锐，并且能给出有用的建议。

大小 7KB 左右
包含 karpathy guide 和 grill-me 的汉化版。
包含对用户隐私信息读取的禁止命令
包含 subagent 使用规范
包含使用 git 和访问数据库的行为规范

建议设置上下文在 150-250K 之间，并安装记忆管理插件。

LLM Wiki

A pattern for building personal knowledge bases using LLMs.

This is an idea file, it is designed to be copy pasted to your own LLM Agent (e.g. OpenAI Codex, Claude Code, OpenCode / Pi, or etc.). Its goal is to communicate the high level idea, but your agent will build out the specifics in collaboration with you.

The core idea

Most people's experience with LLMs and documents looks like RAG: you upload a collection of files, the LLM retrieves relevant chunks at query time, and generates an answer. This works, but the LLM is rediscovering knowledge from scratch on every question. There's no accumulation. Ask a subtle question that requires synthesizing five documents, and the LLM has to find and piece together the relevant fragments every time. Nothing is built up. NotebookLM, ChatGPT file uploads, and most RAG systems work this way.

name	japanese-tech-writing
description	日本語の技術文書・書籍原稿の文章規範。段落と論証の構成（パラグラフライティング）、論証の厳密さ（ツッコミどころの除去）、読み手の負荷の管理、視点と語り、演出の抑制、LLM っぽい空句の禁止、翻訳調の比喩と擬人化の禁止（「運ぶ」「効く」「開かれた問い」など）、冗長の排除を定める。日本語で技術書の章、草稿、記事、解説文を書くとき、または推敲・リライトするときに使用する。
license	Unlicense（https://gist.github.com/k16shikano/67625f2a7d96e3bbdfae8d571a936063）

日本語技術文書の文章規範

日本語で技術的な原稿（書籍の章、記事、解説文）を書く・推敲するときは、以下の規範に従う。

	shader_type canvas_item;

	const int PALETTE_SIZE = 4;

	// CHANGE AT LEAST THESE TWO VALUES IN THE EDITOR!
	uniform vec4 palette[PALETTE_SIZE] : source_color;
	uniform vec4 highlight_color: source_color;

	uniform float scroll = 0.0;
	uniform float scroll_speed = 0.2;

	#lang racket
	;;;
	;;; How to take a screenshot on OS X
	;;; -- Jens Axel Søgaard
	;;;
	;;; This gist illustrates how to work with OS X
	;;; system calls in order to take a screen shot.
	;;;

	(require racket/draw