Awesome Short Drama Skills: 83 short-drama agent skills, and blind tests against a no-skill control

The list went up on Oct 13 and passed 1,000 GitHub stars after @GitHub_Daily featured it on Oct 24 (505 likes, 650 bookmarks). @Ryrenz and @wlzh followed on Oct 25. It sorts 83 skill projects and 16 open-source platforms by production stage: adaptation, episode outlines, hooks and paywall points, script review, storyboards, character assets and video-model prompts. Of the projects that name a target model, the maintainer counts 33 for Seedance, 14 for MiniMax H3 and 6 for Kling.

Most curated lists stop there. This one ran tests. In round 1 (Oct 18), 10 screenwriting skills and a no-skill control each wrote the same brief three times on claude-opus-5-5, and Opus and GPT judged the results blind. Per the maintainer, four skills beat the control in all three runs on both script and character bible: zenstory-ai/drama-skills, Shanyin-ai/shanyin-screenwriting-master, jtydhr88/screenwriting-skills and adoria111/emma-screenwriting-skills. In round 2 (Oct 20), on choosing a premise, both judges put Chinese-WebNovel-Skill, webnovel-writer and Hao0321/ai-short-drama in the top three, and both ranked the no-skill control 9th.

The README is upfront about the limits. Only one writing model was tested. The judges disagreed on 63 of 198 pairs. On episode-ending hooks, Opus often preferred the control. And 'checked by hand' means the maintainer reviewed each repo, not that its output was tested. @Ryrenz also points out that a CC0 list doesn't make the listed skills CC0. Some have no licence, or NC/ND or Elastic terms, so read each one before using it on paid work.

Why it's on ReelStack: it's a working template for the head-to-head skill comparisons this space needs, with a control, repeated runs, two blind judges and published per-skill notes. The list itself is also a skill. npx skills add UnderleveledBuilder/awesome-short-drama-skills installs a finder you can ask which skill fits your workflow.

Our take: start with the four round-1 winners if you write episodic scripts with an agent. Copy the eval setup before you trust any skill's README claims.