op7418 / humanizer-zh

It doesn't call an API or run code — the skill walks the model through spotting and fixing 24 patterns drawn from Wikipedia's "Signs of AI writing" project, grouped into four families: content (over-explaining significance, "-ing"-ending shallow analysis, promotional language, vague attribution), language and grammar (overused "AI vocabulary," negative parallelism, the rule-of-three), style (em-dash overuse, bold overuse, curly quotes), and communication filler (collaborative asides, knowledge-cutoff disclaimers, a sycophantic tone). The README's own before-and-after pairs are the whole demonstration: hand it a paragraph of stiff, AI-flavored Chinese and it hands back something shorter and plainer.

It's for anyone editing or reviewing AI-generated Chinese text — marketing copy, academic summaries, blog drafts — who wants it to read like a person wrote it. Nothing beyond the skill itself is required: no account, no API key, no external service, matching its own recorded cost of free and ready to use.

READ THE FULL ANALYSIS＋

This is a translation and an adaptation, not original research, and it says so. The README states plainly that its core file is translated from blader/humanizer, and that its practical checklist and scoring section were adapted from hardikpandya/stop-slop. Anyone who wants the English original, or wants to check a claim against Wikipedia's actual "Signs of AI writing" page, has to go to those sources — nothing here was independently produced by this repository.