Shanyin Screenwriting Master: four screenplay formats in one prompt file

Four formats from one prompt file, no code to run

Shanyin-ai/shanyin-screenwriting-master is not a program with an entry point. It is a prompt asset: one `.skill` bundle for Claude and GPT, plus one markdown file for chat tools that have no skills module. Nothing in the repository is executable, and there is no install script, no package manifest and no service to start, so adoption costs a download plus a click in a settings panel. The author, @山音, presents himself as an AIGC artist, independent director, screenwriter and writer, and the coverage is organised around production reality rather than genre alone. Four formats are named: concept shorts of 1-3 minutes in either a how-to-tell or a what-if direction, narrative shorts of 5-10 minutes, 90 minute features split into commercial and artistic tracks, and multi-episode series covering a full-season outline, single-episode scripts and character growth lines. That spread suits a writer who wants one deliverable per project type, from a festival entry to a weekly broadcast slot, rather than a single fixed script shape.

The four stage pipeline: character, outline, scene, script

This is the project's real mechanism, and it is a sequence of documents rather than an algorithm. Character design comes first, asking for a personality, a motivation, a growth arc and at least one contrast trait per role. The structure outline stage then picks a classical model, named as the three-act structure or the hero's journey, and pairs it with a genre frame drawn from mystery, science fiction, romance or crime. Scene breakdown follows at storyboard level, where the plan carries camera language hints, emotion nodes and transition design instead of a plain beat sheet. The last stage emits a script in standard film format with scene numbering, action description, dialogue and transitions. Two ordering choices carry weight here: the outline is chosen from a short fixed menu before any prose is drafted, and the target format is settled before the stages begin, so a 90 minute commercial feature and a 1-3 minute concept short travel the same four steps with very different length targets.

Importing into Claude and triggering with plain sentences

Installation means two clicks in a client and one file picker. The repository holds four entries at its top level, and only two of them are the skill itself:

LICENSE
README.md
screenwriting-master（Claude&GPT通用）.skill
山音超级编剧大师集成版（Gemini及其他AI工具通用）.md

For Claude, the documented path is a three step walk through the desktop app or the web version, and finishing the import is also what enables the skill:

打开 Claude App / 网页版 → 进入「Skills」模块 → 点击「Import Skill」

Point the dialog at the `.skill` file you downloaded and the skill becomes available to enable. Nothing else has to be configured: no API key, no environment variable, no model identifier to pin, no settings file to edit. First use is shorter than that, because the trigger is ordinary language with no required command or keyword, and any content can be interacted with once the skill is loaded. The cost of that convenience shows up in control: with no documented trigger phrase, there is no stated way to force one specific stage, so naming the format and the stage inside your first sentence is the only routing the project exposes.

Audience, style tags and platform length as tuning knobs

Advanced use adds detail rather than configuration. Three kinds of supplement can be handed to the model: an audience, named as teenagers, an adult audience or an overseas market; a style tag, named as Wes Anderson texture, Nolan suspense or Miyazaki healing; and a platform constraint, named as short video platform rhythm, theatrical running time or the weekly logic of a series. None of these are files or flags, they are sentences appended to the conversation, so two writers using the same imported skill can end up with different documents without any settings diverging on disk. The platform constraint has real weight against the four format claim, because a theatrical feature and a short video cut are different products rather than two lengths of one script. What the project does not supply is any check on that input, since the README sets no word count per format, no minimum beat count and no fallback for a brief so vague that it names neither genre nor length.

What the repository ships contradicts what the README promises

What sits in the repository is not always what the README calls it. The core file is described as 山音超级编剧大师.skill, yet the top level listing carries `screenwriting-master（Claude&GPT通用）.skill`, so the name you are told to look for is not the name on disk. That same filename advertises itself as usable in Claude and GPT, and the walkthrough that follows covers Claude alone, with no menu, screen or file picker given for any GPT product. Second, the README offers `examples/` as optional case material, naming a science fiction concept short, an artistic feature outline and an urban series script, but no `examples/` entry appears among the four top level files, so those samples cannot be read from a clone. Conditions travel with the file as well, and they are not a stock grant: the README opens by declaring MIT and then attaches four conditions of its own, two of which narrow what that grant normally allows. Free download, import and use for personal or commercial script writing come first, and adjusting the screenwriting logic inside the file comes second. The third condition keeps the @山音 credit in place, and the fourth bars reselling the file, bundling it commercially or shipping it as a core component of a paid product. Read together, the output is free and the tool is not, which is a distinction worth settling before a studio adopts it.

Nothing is documented about updating or rolling back a skill

What happens after the import is where the documentation thins out. The last push was on 2026-04-01 and the project publishes no versioned build, so there is no upgrade note to read, no changelog to diff and no way to tell whether the two shipped files changed after you installed them. The README also stays silent on removing a skill from a client, on pinning a model version, and on what to do when the underlying model returns a script that quietly ignores the requested format. None of that is unusual for a collection of prompts, but it pushes the quality question onto your own testing: the repository ships no evaluation suite, no reference output and no accuracy figure, so the only honest check is to run one brief through all four stages twice and compare what comes back. A second round with the same brief and a different model is the cheapest test of whether the format instruction carried any weight.

Two ways to load the same instructions, different control

Two distribution shapes ship for the same instructions, and they differ in where the text ends up. The `.skill` bundle is loaded through the client's skills module and enabled from a settings panel, so the client is what manages it. The integrated markdown file goes the other way:

山音超级编剧大师集成版（Gemini及其他AI工具通用）.md

You drag that file into the chat box, and from that moment it is plain text inside a single conversation, with no list entry and nothing to switch off. Gemini is named as the reason this second file exists, and the drag is the entire procedure. For a writer planning many scripts in one client, the bundle removes a copy and paste step from every conversation and keeps the same four stage order available. For one script with a house style, pasting the markdown file once and adding audience and style details in ordinary sentences is enough, and nothing needs to be installed, configured or enabled first. That second path is also the only one that reaches a tool with no skills module at all.

The pipeline stops at the finished script

The four format claim is also where the project runs out of road. Its pipeline ends at the finished script: scene breakdown covers camera language hints, emotion nodes and transitions, and nothing beyond that level appears anywhere in the documentation, so a shot list, a storyboard, a shooting schedule, a budget or a production contract sits outside its scope. Broad format coverage also does not mean broad genre expertise. Mystery, science fiction, romance and crime appear as frames to pair with a chosen model, not as genre playbooks carrying their own conventions, and the two classical models it names are offered as choices rather than as structures worked through beat by beat. If your real bottleneck is one genre's rules, one studio's script format or one director's habits, this file hands you a scaffold and then stops, leaving the conventions to be written by whoever imports it.

Editorial conclusion

Adopt this skill if you write inside Claude, GPT or Gemini and want one consistent four stage path from character notes to a formatted script, with the format chosen before drafting begins. Leave it if you need a shot list, a shooting schedule, a studio format template or a genre playbook, because the pipeline stops at the finished script and the genre frames are names rather than rules. Before importing, confirm your client accepts the shipped bundle, check that the four files you cloned match the two you expected, and settle the four conditions attached to the file before any commercial script starts.

Frequently asked questions

Official sources

1. Issues
2. License: MIT
3. README
4. Shanyin-ai/shanyin-screenwriting-master on GitHub