VanGong1999/screen-creative-skills: 31 Agent Skills for Film and TV Script Work

The screening bottleneck these skills target

The repository is aimed at a specific job: reading a large volume of prose and turning it into a decision. Its README frames the audience as 影视创作者、策划师、内容评估师, that is, film and television creators, planners and content evaluators, and lists familiar situations such as a web-novel platform screening more than 100 submissions a day, a studio assessing its IP library, or an individual writer hunting for adaptable source material. The 31 skills are grouped into six families: four for novel screening, eight for story analysis, two for character work, four for evaluation, two for creation planning, and eleven auxiliary skills covering workflow orchestration, result handling, tooling and knowledge retrieval.

What makes the grouping coherent is that each family maps to a stage of the same pipeline. A novel goes in, a summary and a score come out, then plot points, then characters, then an outline, then an evaluation of that outline. The README does not describe a single entry point that runs all of this; it describes skills that are called individually, and it states that every stage can be invoked on its own so the user keeps control of direction. That is a reasonable design for evaluation work, where a human usually wants to stop and argue with the machine after each pass.

The claim that this yields a tenfold efficiency gain appears in the README without a described method. No benchmark, dataset or timing is given. Treat it as a marketing line, not a measurement.

How the skill families chain together

The mechanism visible in the repository is a directory of skill definitions plus supporting documentation, not a running service. The top level contains LICENSE, README.md, category/, docs/, english-version/ and two QR code images. The category/ directory is where the six families live, and docs/ holds the longer explanations. There is no package manifest at the top level, which tells you the Python language tag reflects the code that ships inside individual skills rather than a single installable application.

The data flow implied by the README runs in one direction. novel-summarizer condenses a manuscript into a 500 to 800 character outline while preserving character relationships. novel-evaluator scores the text across multiple dimensions and reports adaptation potential. score-analyzer aggregates scores from several evaluation rounds and computes a rating grade. From there, story-five-elements produces a five-axis analysis of genre, characters, relationships, plot and originality, and generates a mind map. drama-analyzer extracts plot points and classifies their dramatic function, aimed at medium-length and long series. plot-keypoints identifies three-act structure and core turning points; detailed-plot-analyzer expands those into a finer breakdown with an emotional curve.

The two character skills sit alongside rather than inside that chain. character-profile builds a multi-dimensional portrait with background and motivation, and character-relationships maps relationship types and how they evolve. Both are marked as suitable for all content types, which is the honest answer: character work does not depend on episode length, whereas plot-point extraction does, and the README reflects that by restricting drama-analyzer and detailed-plot-analyzer to longer formats.

Installing screen-creative-skills and running a first novel screening

The README does not document an installation procedure. There is no pip command, no requirements file at the top level, no environment variable and no port. The repository lists no homepage either. The only pointers are three sibling repositories named in the README: a script knowledge base, a second copy of the screen creation skills, and a repository for film and television creation agents. If you want to use these skills, the realistic path is to clone the repository and read the skill definitions, then paste the relevant one into whichever agent runtime you already use.

The commands below reflect that reading path. Cloning is the only step the repository layout clearly supports.

git clone https://github.com/VanGong1999/screen-creative-skills.git
cd screen-creative-skills
ls category/ docs/

After that you should see the six skill families under category/ and the longer documentation under docs/. If you are working in a non-Chinese agent stack, the english-version/ directory is the one to open first, since the README itself is written in Chinese and the skill descriptions follow the same language.

A first real use looks like this: pick novel-summarizer, open its definition, and feed it a manuscript section. The README states the expected output is a 500 to 800 character outline that keeps the narrative intact and surfaces character relationships. If your draft comes back at 200 characters or loses a subplot, the skill definition is where you adjust the instruction, not a config file, because no config file is documented.

ls category/
cat docs/README.md

The second command is worth running before you trust any of the output formats, because the README leaves the scoring rubric for novel-evaluator and script-evaluator entirely unspecified.

Where the documentation stops and guesswork starts

The largest limitation is that the evaluation criteria are never published. script-evaluator is described as assessing a script from three angles, translated in the README as ideological, artistic and appreciative quality, and as marking strengths and weaknesses and producing improvement suggestions. What the scale is, how many points, how the three dimensions are weighted, and what separates a passing script from a failing one are not in the README. score-analyzer computes a rating grade from multiple rounds of scores, but the mapping from score to grade is likewise absent. For a tool whose entire purpose is to support a go or no-go decision, that is the gap that matters most.

The second limitation is language. Every skill name is English, but the descriptions, the expected outputs and the example scenarios are Chinese. The english-version/ directory exists, and the README does not state how complete it is. If your team writes coverage in English, verify the translation coverage before planning around it.

The third is the absence of releases. No releases were retrieved for this repository, and the README carries a version number of 2.1.0 with an update date of 2026-01-11. That means versioning is manual and there is no changelog to read when behaviour shifts between pulls. The last push to the default branch was on 2026-05-26, so the project has moved since that README date, and you cannot tell from the repository metadata what changed.

Finally, this is the wrong tool if you want a service. There is no API, no server, no batch runner described. It is a set of instructions for an agent that you supply.

How it differs from a general-purpose writing prompt set

The obvious alternative is a general prompt collection for creative writing, or simply writing your own prompts for a chat model. The difference in approach is the decomposition. A general writing prompt asks for a treatment or a scene and returns prose. These skills are built around extraction and judgement instead: novel-summarizer produces a fixed-length outline, plot-keypoints produces a structural skeleton, character-relationships produces a network rather than a description. The output is meant to be compared across submissions, which is why score-analyzer exists at all. A generic prompt set has no equivalent of aggregating several evaluation rounds into a grade.

The cost of that decomposition is rigidity. A general prompt adapts to a strange manuscript; a skill with a stated 500 to 800 character target will fight you if the source is a 40-episode series. The README handles format variation by restricting skills to content types, with drama-analyzer and detailed-plot-analyzer limited to medium-length and long series, and drama-evaluator limited to vertical short drama. That is a sensible boundary, but it also means you cannot take a short-drama manuscript through the long-series plot analysis without expecting poor output.

A second alternative is a dedicated script coverage tool with a published rubric. Those give you a defensible score but rarely let you inspect or edit the criteria. This repository gives you the opposite trade: everything is editable text, nothing is calibrated.

Licence, maintenance and what an upgrade costs

The repository is MIT licensed. For a collection of prompt and skill definitions, that is the permissive end of the spectrum: you can copy the files into a commercial pipeline, modify them and redistribute them, provided the copyright notice and permission notice are retained. This is a description of the licence terms, not legal advice; if you are folding these definitions into a product, have your own counsel read the LICENSE file at the top level rather than relying on the one-line summary here.

The practical licence question is different. If you fork a skill and tune its scoring language for your own slate, your modifications are yours to keep private under MIT, but they also diverge from upstream with no changelog to reconcile against. With no releases and no tags retrieved, upgrades are pull-based: you compare the files you changed against the new versions by hand. For a repository of this size, that is manageable, but it is a real recurring cost if you have customised a dozen skills.

Maintenance should be judged from the push date alone. The last push was on 2026-05-26. The README's own version marker is 2.1.0 dated 2026-01-11, so the documentation lags the code by roughly four months. Nothing in the repository metadata indicates an archived state.

Editorial conclusion

Adopt it if you work in Chinese-language film and television development and want reusable, inspectable prompts for novel screening, plot-point extraction, character profiles and script scoring; the MIT licence and the plain skill layout make it cheap to copy and adapt. Do not adopt it if you need a runnable pipeline, English-language output, or published evaluation criteria, because the README states no install command, no dependency list and no scoring rubric. Before committing, open the category/ and docs/ directories and read two or three skill files to confirm the output format matches the reports your team already produces.

Frequently asked questions

Official sources

1. Issues
2. License: MIT
3. README
4. VanGong1999/screen-creative-skills on GitHub