An awesome list that actually tests the skills it recommends

What it does

It catalogs 83 agent skills (plus 16 related open-source platforms) for making vertical short drama and AI manju with agents: novel adaptation, episode outlines, hooks and paywall cliffhangers, script review, storyboards, and video prompts for Seedance, Kling, and MiniMax. Most entries are standard SKILL.md format, so they load into Claude Code, Codex, Qwen, Doubao, Kimi, and friends. Everything is human-checked and sortable by type, production stage, and target video model.

The interesting bit

Unlike most awesome lists, this one runs actual experiments. Two rounds of blind head-to-head evals — judged by Opus and GPT, each skill writing the same prompt three times against a no-skill control — produced ranked winners for screenwriting and topic-picking, with per-skill notes like “hooks end too softly” and “paywall points are weak.” The list even documents its own limitations: only one model tested, judges disagreed on 63 of 198 pairings, and the control group beat three of the skills.

Key highlights

• Two published eval rounds with named winners, per-skill strengths and weaknesses, and honest failure notes
• Recommended multi-skill “packs” for common pipelines (novel-to-storyboard, Seedance output, long-running manju serialization)
• The list itself ships as a skill — short-drama-skill-finder — so your agent can pick skills for you
• Entries flagged for staleness, archival, dead repos, and non-standard prompt files; suspicious file reuse between repos is called out
• Companion site at shortdramaskills.com with full eval writeups

Caveats

• The recommended packs are assembled from repo descriptions, not tested — the README says so upfront
• “Human-checked” means verifying the repo, structure, and categorization, not measuring output quality
• Eval drafts came with numbering and self-check tables baked in that must be stripped before handing off to a video model

Verdict

Worth a look if you’re building AI short-drama pipelines and want evidence instead of star counts. If you’re not in the vertical-drama or manju space, it’s a well-executed list for a niche you don’t live in.