What is no-ai-slop-zh-tw?

This repository is a Traditional Chinese adaptation of an English project called "no-ai-slop." It is an Agent Skill, meaning a set of instructions you can plug into AI coding tools like Claude Code or Codex. The skill edits Traditional Chinese writing to remove common AI phrasing patterns while keeping the original author's voice, rhythm, and personality. It can also work in a detection-only mode that points out problems without rewriting anything, and it does not claim to determine whether text was actually written by AI. The English original focused on patterns specific to English writing. This version redesigns the rules for Traditional Chinese as used in Taiwan. It replaces English-specific patterns, like sentences ending in "-ing" or overuse of dashes, with Chinese equivalents: empty analysis phrases like "demonstrates commitment to innovation," overuse of dashes, stacked conjunctions, and mechanical parallelism. It also adds detection for patterns common in Chinese bureaucratic writing, nominalization, false completeness, and causal claims made without evidence. The skill targets more than twenty patterns. Examples include binary pivot templates ("the real issue is not X, but Y"), throat-clearing openings ("it is worth mentioning that..."), fake insight setups ("most people overlook..."), vague attribution ("research shows," "experts generally agree"), and dramatic fragmented sentences. It also catches mechanical parallelism like "faster, more accurate, smarter" and paragraphs that always begin with connectors like "additionally" or "meanwhile." Beyond pattern removal, it applies basic editing principles: leading with the point, using direct verbs, breaking up hard-to-read sentences, and preserving concrete details. To install it, you fork the repository to your own GitHub account, then paste a single instruction into your AI tool telling it to install the skill globally. Usage involves calling the skill with a slash command followed by your draft. In editing mode, it returns a revised draft plus a short summary of changes. In detection mode, it lists each pattern found, quotes the relevant sentence, and suggests a fix direction without rewriting. The repository includes six files: the main rules and workflow, a post-editing checklist, examples of edits and things that should not be changed, an OpenAI configuration file, an upstream tracking document, and a fork guide. It is released under the MIT license.