Short drama skill tests
"Checked" in the directory only means a project's details are accurate, not that it works well. This page holds the tests: the same premise goes to different skills, and the outputs are laid side by side unedited so you can judge for yourself.
Status: round one, covering script-writing skills, is in preparation. Results will be posted here.
Method
- Same input: one premise with genre, episode count and platform, shared by every skill in the round.
- Same environment: the same agent, model and reasoning effort, with versions recorded.
- Pinned skill versions: the commit of each repo at test time is recorded.
- Each skill's own process: if a skill asks questions, we answer them honestly and add no extra instructions; the whole exchange is recorded.
- Every output kept: no cherry-picking; every run is published.
What we look at
| Dimension | What we look for |
|---|---|
| Hooks | Do the opening and each episode ending make you want to keep watching? |
| Characters | Are motives clear and behavior consistent? |
| Structure | Do pacing, escalation and twists hold up? |
| Dialogue | Does it sound like people talking, and do characters sound different? |
| AI tells | How many stock phrases, tricolons and empty adjectives? |
| Usability | How much editing before it is usable, and how smooth is the process? |
Reviews give observations and examples, not overall scores. Raw outputs and logs go in the evals folder on GitHub, so anyone can reproduce the method.
Want a skill included?
Open an issue on GitHub with the repo link and the genres it is good at.
Updated 2026-10-17