How to make an entire GitHub account's worth of repos findable — by GitHub search, Google, and AI assistants — and how to do it at scale with a multi-agent workflow. Distilled from polishing 110+ public repos in one pass.
| #!/usr/bin/env bash | |
| # | |
| # ultrathink.sh — recreate Claude Code's "ultrathink" rainbow shimmer animation | |
| # | |
| # How the real thing works (reverse-engineered from the Claude Code v2.1.175 bundle): | |
| # | |
| # 1. The input editor matches /\bultrathink\b/gi on your prompt text. | |
| # 2. Each character i of the match gets a static color: palette[i % 7], | |
| # a 7-color ROYGBIV palette defined in the dark theme, rendered bold | |
| # with 24-bit ANSI escapes (\e[38;2;R;G;Bm). |
A curated, heuristic guide distilled from actually building and publishing one (anki-openevidence-addon, AnkiWeb 43671435). It focuses on the parts that are undocumented or surprising — especially automating AnkiWeb uploads, which has no official API.
Distilled from the iteration history of a hands-on Quarto workbook — a half-day workshop teaching clinical staff (no prior coding experience) to reproduce a published paper's analyses in R, using a free-tier LLM (ChatGPT / Claude / Gemini) as their coding assistant. The lessons are deliberately written topic-agnostic so they transfer to any "non-coders + LLM + data" workshop.
The meta-lesson first: almost every commit in this project was a friction point discovered in practice, not predicted in advance. You cannot armchair your way to a newbie-friendly workshop. Ship a v1, watch where clinical staff actually trip, and remove that exact stone. The changelog is the curriculum-design document. Treat "where did a learner get stuck for >2 min?" as your only real backlog.
| # polish-prompt | |
| # Run `make help` for the menu. | |
| ROOT := $(abspath $(dir $(lastword $(MAKEFILE_LIST)))) | |
| LABEL := com.example.polish-prompt | |
| PLIST := $(ROOT)/$(LABEL).plist | |
| LAUNCHAGENT := $(HOME)/Library/LaunchAgents/$(LABEL).plist | |
| DOMAIN := gui/$(shell id -u) | |
| ENV_FILE := $(HOME)/.config/polish-prompt/env |
把常用的 Todoist 操作包成 Claude Code 的 global skill,從此在任何專案裡對 Claude 說一句中文,他就能幫你 CRUD 待辦事項。
- 做了什麼:在
~/.claude/skills/todoist/下打造一個 global skill,包住 Todoist REST API v1 的 CRUD。 - 怎麼觸發:對 Claude 說「加個『買牛奶』明天早上九點到 Todoist」、「看我今天待辦」、「把運動那件事改到下午五點」——skill 的
description已經布好中文觸發詞,Claude 會自己載入。
住院醫師的移植後發燒處理無腦流程圖,依時間和抗生素使用狀況分階段處理
NCCN Guidelines® References (v1.2026):
- FEV-5: Initial Inpatient Empiric Therapy for Fever and Neutropenia (p. 33)
- FEV-A-2: Antibacterial Agents - Anti-Pseudomonal (p. 42)
- INF-1: Antimicrobial Prophylaxis Based on Overall Infection Risk (p. 7)
- FEV-8: Evaluation and Treatment Modifications for Specific Clinical Presentations — Day 0 (p. 36)
A practical reference for wiring up end-to-end mail triage in Claude Code. What you get at the end:
- One natural-language command ("process my mail" / "24 小時 actionable") triages Gmail personal + Gmail work + enterprise Outlook in a single pass.
- Enterprise Outlook works even if your tenant has disabled OAuth consent (
AADSTS65002/need admin approval) — scraping OWA via a stealth Chromium session instead of IMAP/Graph. - Trigger it from your phone via a Telegram bridge. Claude runs headless in
tmuxon your Mac; you text from bed, a prioritized digest comes back. - Read-only Outlook tenant is still actionable — instead of moving items, Claude creates macOS Reminders and keeps an append-only action log as the only source of truth.
Who this is for. You use Claude Code as your daily driver, you're comfortable with shell + Python, and at least one of your accounts is on a