0xsline/short-drama: A Short-Drama Script Creation Skill

What short-drama solves

0xsline/short-drama targets writers, editors, and AI-assisted production teams that need a repeatable process for creating short-form episodic drama. Its scope is broader than a single prompt. The documented workflow starts with project setup, moves into a story plan and character system, creates an episode directory, writes individual episodes, reviews them, checks compliance, and finally exports the accumulated material. That sequence addresses a common coordination problem: long-running story development can lose consistency when prompts, notes, drafts, and reviews are kept separately.

The skill is designed for both Chinese domestic formats and overseas English formats. Domestic output uses scene headings, music prompts, camera notation such as △, and dialogue lines. Overseas mode changes those conventions to headings such as `INT. LIVING ROOM - DAY`, `WIDE SHOT`, and English character labels. The project therefore serves writers who need a guided structure, but it also makes assumptions about short-drama conventions, including hook placement, paywall markers, episode counts, and a four-stage rhythm curve.

The intended users are people working with AI coding assistants that can load skills from a designated directory. It is not a standalone screenplay application described in the repository. There is no graphical interface, database, media renderer, or model-independent quality guarantee in the documented project structure. Its value comes from organizing the assistant's work around named commands and files.

Commands and project state

The architecture is file based. A new project is configured with `/start`, which collects genre, audience, tone, ending type, episode count, language, and mode. The README states that this creates `.drama-state.json`, and later commands read that state. The documented state example includes the current step, selected genres, audience, tone, total episode count, completed episodes, language, mode, and title. This gives an assistant a compact project summary, but the repository does not explain whether the state file is versioned, validated against a schema, or protected against conflicting concurrent edits.

The command sequence has explicit prerequisites. `/plan` follows `/start` and creates `creative-plan.md`. `/characters` follows `/plan` and creates `characters.md`. `/outline` follows `/characters` and creates `episode-directory.md`. `/episode {N}` requires the outline and writes files under `episodes/`. The `/review {N}` command produces a review under `reviews/`, while `/compliance` writes `compliance-report.md`. `/export` requires at least some written episodes and creates an `export/` directory.

The state mechanism is useful for continuity, yet it is also a dependency. If the assistant cannot read `.drama-state.json`, the documented command prerequisites cannot be confirmed. A team should inspect the generated files after each stage rather than assuming that a completed command means the underlying story decisions are correct.

Installing the skill and creating a first project

The README documents both global and project-level installation. A global installation places the repository in the assistant's skills directory. The project-level option places it under `.claude/skills/short-drama` inside a specific project. These commands use the repository URL exactly as shown:

git clone https://github.com/0xsline/short-drama.git ~/.claude/skills/short-drama

The global command targets macOS or Linux. For Windows with Git Bash, the README gives this command:

git clone https://github.com/0xsline/short-drama.git "$USERPROFILE/.claude/skills/short-drama"

For a project-local installation, clone the repository into the project's `.claude/skills/short-drama` directory:

cd your-project
git clone https://github.com/0xsline/short-drama.git .claude/skills/short-drama

After installation, the README says to start the AI coding assistant and enter `/start`. If the topic selection interface appears, installation is considered successful. The first real creation flow can then follow the documented stages. The README uses these command examples:

/start
/plan
/characters
/outline
/episode 1
/review 1

The sequence produces project state, a creative plan, character documentation, an episode directory, the first episode, and a review report. The repository does not document a dependency installation step, environment variables, ports, or a test command.

Episode writing and overseas formatting

The `/outline` stage turns the larger plan into an episode table. Each entry is documented as containing an episode number, title, one-sentence summary, hook type, and marker. The outline also tracks hook distribution and rhythm segments. The skill defines five hook categories: suspense, reversal, emotion, information, and crisis. It gives reference ranges for each category, including 30 to 40 percent for emotional hooks and 5 to 15 percent for reversal hooks. Those are design rules in the README, not evidence that every story will perform well under them.

Episode files use a predictable Markdown structure. A domestic episode begins with metadata about its segment, hook, marker, duration reference, and emotional curve, followed by two to five scenes. The format supports music, camera descriptions, and character dialogue. This is useful for keeping generated drafts readable, but the repository does not specify a production script interchange format such as Fountain, Final Draft, or a video editor timeline.

The `/overseas` command switches the draft to English and a Hollywood-style notation. The README gives this direct example of the change in presentation:

## Scene 1: INT. LIVING ROOM - DAY
WIDE SHOT: A room description
**SU NIAN**: (smiling) Hello

The documented mode also adjusts names, scene descriptions, pacing, and culturally sensitive material. That last promise should be read as an assistant instruction, not a guarantee of successful localization. The repository does not identify a translation model, target-country rule set, or human review requirement for overseas output.

Review scoring and compliance boundaries

The `/review` command applies five dimensions, each scored from 1 to 10, for a maximum of 50. The dimensions are pacing, satisfaction, dialogue, format, and continuity. The README maps score bands to editorial states: 45 to 50 is described as directly usable for filming, 38 to 44 as requiring minor adjustment, 30 to 37 as requiring targeted changes, 25 to 29 as needing structural improvement, and below 25 as requiring rewriting. The report is supposed to include detailed comments, retained strengths, prioritized issues, an ASCII emotion waveform, and revision order.

This scoring system creates a consistent review artifact, but it does not define the rubric implementation beyond the listed criteria. It also cannot establish factual correctness, production feasibility, audience response, or compliance with a particular distributor's rules. A high score in the documented scale means the assistant applied its criteria, not that a screenplay has cleared every external review.

The `/compliance` command is separate and checks political sensitivity, beautification of illegal or criminal behavior, sexual or violent content, discrimination and insults, risks involving minors, religion, ethnicity, historical events, medicine and law, social influence, educational meaning, and advertising placement. Its output is `compliance-report.md`. The repository does not state which jurisdictions the checklist covers, whether platform-specific rules are loaded, or who resolves flagged content. That makes compliance output a triage aid rather than final approval.

Strengths, limitations, and the practical boundary

The clearest advantage is the explicit production chain. A writer can begin with `/start`, establish a plan and character system, outline episodes, draft scripts, review individual files, and export the accumulated package. The file layout also makes artifacts easy to locate: plans, character documents, episodes, reviews, compliance reports, and exports are separated by name. Built-in references cover genres, openings, rhythm, hooks, paywalls, satisfaction patterns, villains, and compliance, which gives the assistant a defined vocabulary instead of relying on one generic writing prompt.

The main limitation is that the system appears to coordinate instructions and Markdown files rather than control an external production environment. The README does not document API calls, editor integrations, rendering, character database synchronization, collaborative conflict resolution, or rollback. It also does not define how the assistant handles contradictory changes when a user revises the premise after outlining several episodes. The hook ratios, 10 to 15 percent paywall guidance, and four-stage pacing curve are useful prompts, but they can flatten stories that need a different rhythm.

A real alternative is to use a general-purpose screenplay-writing assistant with manually maintained Markdown files. That approach offers more freedom over prompts and output structure, but it lacks this repository's named command sequence, state file, review dimensions, and bundled reference documents. Another alternative is a production suite that imports structured screenplay formats and manages revisions. The repository does not name such a product, and it does not claim compatibility with one. The practical distinction is control: short-drama supplies conventions and workflow, while a general assistant or production suite may supply more explicit integrations.

Editorial conclusion

Adopt 0xsline/short-drama when a team needs a structured AI-assisted workflow with domestic or overseas formatting, predictable files, and built-in review stages. Do not treat it as an autonomous production system or as a substitute for platform policy review. Verify first that the assistant supports the installed skill directory and that its generated scripts meet the intended market's editorial, legal, and production requirements. The repository uses the MIT license, and its last push was on 2026-04-17.

Frequently asked questions

Official sources

1. 0xsline/short-drama on GitHub
2. Issues
3. License: MIT
4. README