chatgpt-comparison-detection: Corpus First, Then Detection
Dissecting an early AI text detection repo: its most valuable part is the HC3 human vs. ChatGPT comparison corpus.
AI Skill breakdowns
Built from real files: how a Skill works, how it is written, and what is worth reusing.
Notes
Dissecting an early AI text detection repo: its most valuable part is the HC3 human vs. ChatGPT comparison corpus.
Breaking down blader/humanizer: a set of English writing review rules designed for AI.
Comparing Humanizer-zh and the English humanizer: They help you break down AI writing traces into specific issues.
A deep dive into shuorenhua: it first protects facts, then suppresses the AI tone in Chinese.
A deep dive into stop-slop: a writing checklist with no scripts, only hard rules.
A deep dive into yzhao062/agent-style: it uses 21 rules, adapters, and a style-review Skill to help AI avoid writing templated, robotic technical documents.
A deep dive into alchaincyf/nuwa-skill: what’s truly worth learning is how it incorporates research, refinement, and boundaries into a Skill.
A deep dive into Anbeeld/WRITING.md: a set of rules that make AI writing more context-appropriate and less templated.
A deep dive into jzOcb/writing-style-skill: it uses the differences between the original and final drafts to gradually distill your writing style back into SKILL.md.
A deep dive into baoyu-article-illustrator: it breaks down image addition into reading the article, deciding placement, writing prompts, and generating images.
A deep dive into baoyu-post-to-wechat: it links Markdown, cover images, content images, and the WeChat Official Account API, saving posts to the draft box first for human review.
A deep dive into baoyu-url-to-markdown: it collects webpages, posts, and video subtitles locally, turning them into materials you can directly use for later writing.
A deep dive into taste-skill: a Skill dedicated to handling the 'flavor' of AI-generated frontend page templates.
A deep dive into nutlope/hallmark: instead of just saying “make it look sophisticated,” it uses macro structures, themes, anti-template checklists, and pre-delivery audits to force AI away from default patterns.
A deep dive into alchaincyf/huashu-design: it turns prototypes, PPTs, animations, and infographics into an HTML design workflow, focusing on producing real directions first, then letting people decide.
A deep dive into JuliusBrussee/caveman: it’s not about making AI cute, but about turning “say less fluff” into rules, boundaries, installation adaptations, and statistics.
A deep dive into colbymchenry/codegraph: it builds a code structure graph upfront and then provides it to Agents via MCP.
A deep dive into grill-me: how it turns a vague plan into an actionable process by asking clarifying questions.
A deep dive into GordenSun/GordenSuperPPTSkills: it creates a complete production line by combining image generation, visual layer decomposition, and PPTX recomposition.
A deep dive into op7418/guizang-ppt-skill: it uses HTML, templates, locked layouts, and validation scripts to turn presentation PPTs into a stable visual system.
A deep dive into rediumvex/ai-video-generator-claude: it uses 10 Claude Skills to break down short video prompts into hooks, shots, lighting, sound, and platform adaptation.
A deep dive into bkingfilm/lapian-notes: it extracts frames locally, matches subtitles, packages data for AI analysis, then imports JSON back into timelines, structure trees, and emotion curves.
A deep dive into siuserxiaowei/reference-driven-cinematic-video-skill: it turns product briefs, reference video analysis, voiceover subtitles, and quality gates into a Codex video production workflow.
Deep dive into YILS-LIN/short-video-factory: it uses an Electron desktop app to chain copywriting, TTS, random material segmentation, and FFmpeg into a local production line.
A deep dive into browser-use/video-use: it first converts videos into transcripts and timeline visuals, then lets an Agent generate EDLs, render, and self-check.
How these are written
Each note covers one Skill. Claims stay tied to source files, documentation, or observed use.
The note starts with what gets stuck in real use, not with a generic project tour.
Implementation claims stay tied to source files, rules, scripts, or documented behavior.
The ending turns the case into a practical way to build a similar Skill.