Screenwriting Skills: 24 Agent Skills for Film and TV Writing
Explore 24 reusable agent skills for screenwriting, television, dramaturgy, Chinese opera, and script analysis across Claude Code and Codex.
Screenwriting Skills: 24 Agent Skills for Film and TV Writing
Writing software can generate scenes quickly. Writing a script that sustains conflict, controls information, respects format, and survives revision is much harder. The open-source jtydhr88/screenwriting-skills repository addresses that gap with a structured library of 24 agent skills for screenwriting, television writing, stage drama, and dramaturgy.
The project turns ideas from 45 craft books and 23 published scripts, scores, and plays into reusable SKILL.md files. These files are designed for Claude Code and OpenAI Codex, but they follow the agent-neutral openagentskills.io format, so any compatible agent can potentially use them.
What problem does this solve?
A general-purpose language model may know that a screenplay needs a protagonist, conflict, scenes, and dialogue. It may still produce exposition-heavy conversations, weak act-outs, inconsistent character motivation, or television episodes that have no repeatable story engine.
These skills provide a more disciplined working process. Instead of asking an agent to “make the script better,” a writer can invoke a focused capability for:
- Finding the controlling idea and testing a premise
- Building a feature or series structure
- Designing a character web that generates conflict
- Rewriting dialogue as action rather than information delivery
- Breaking an episode into acts and act-outs
- Creating a series bible and selling documents
- Adapting a novel into a multi-episode drama
- Analyzing published scripts and dramatic traditions
- Maintaining project state across multiple writing sessions
The result is closer to a craft-aware writing assistant than a generic text generator.
Installation
Claude Code marketplace installation
The recommended Claude Code installation uses the plugin marketplace:
/plugin marketplace add jtydhr88/screenwriting-skills
/plugin install screenwriting@screenwriting-skills
Skills are then available with the /screenwriting:<skill> namespace. For example:
/screenwriting:sw-dialogue
To see the skills loaded in the current environment, run:
/skills
Claude Code: install for every project
git clone https://github.com/jtydhr88/screenwriting-skills.git
cp -r screenwriting-skills/plugins/screenwriting/skills/* ~/.claude/skills/
Claude Code: install in one project
mkdir -p .claude/skills
cp -r screenwriting-skills/plugins/screenwriting/skills/* .claude/skills/
Codex installation
For Codex CLI, the ChatGPT desktop app, or an IDE extension, add the marketplace and install the plugin:
codex plugin marketplace add jtydhr88/screenwriting-skills
Then choose Plugins → Screenwriting Skills → Install.
The same 24 SKILL.md files are shipped through the plugin's shared skills/ directory. There is no separate rewritten Codex edition.
For a personal Codex installation:
git clone https://github.com/jtydhr88/screenwriting-skills.git
cp -r screenwriting-skills/plugins/screenwriting/skills/* ~/.agents/skills/
For a project-specific installation:
mkdir -p .agents/skills
cp -r screenwriting-skills/plugins/screenwriting/skills/* .agents/skills/
You can invoke a skill explicitly with a command such as $sw-story-structure, or let the agent detect the relevant context automatically.
A four-layer architecture
The repository organizes its skills into four layers. This matters because a feature screenplay, a streaming series, a stage play, and a Chinese opera may share some dramatic principles while requiring very different structures and writing rules.
1. General dramaturgy
The medium-independent layer covers the foundations used across projects:
sw-workflow: project orchestration, deliverables, stage gates, and persistentstory-bible.mdfilessw-story-structure: plot points, paradigms, beats, complications, crises, climaxes, subplots, openings, and endingssw-premise-theme: premise tests, controlling ideas, loglines, thematic conflict, and the story coresw-character-conflict: character psychology, opposition, orchestration, and escalating conflictsw-dialogue: dialogue as action, subtext, exposition, character vocabulary, and line-level problemssw-scene-craft: value turns, scene analysis, pacing, transitions, props, and entering late or leaving earlysw-format-adaptation: screenplay formatting, Fountain output, Chinese and Japanese script conventions, adaptation, and revision
A useful feature workflow might move from premise to structure, then character, scenes, draft, revision, and submission. The workflow skill names the relevant skill at each stage and defines advisory exit checks.
2. Medium-specific writing
The second layer adds rules that general screenwriting books cannot fully teach:
sw-series-structure: teasers, cold opens, broadcast act grids, streaming act tests, act-outs, cliffhangers, A/B/C story weaving, pilot beats, and season shapesw-series-engine-bible: repeatable series engines, character webs, pilot types, story landmines, and selling documentssw-writers-room: breaking stories, showrunner responsibilities, staff roles, document chains, notes, schedules, and production constraintssw-sitcom-comedy: half-hour comedy, joke mechanics, running gags, cold opens, tags, multi-camera, single-camera, animation, and dramedysw-chinese-series-practice: Chinese television formats, episode suspense, adaptation workflows, production gates, contracts, and content boundariessw-chinese-opera-banqiang: the banqiang vocal system used across traditions including Peking, Yu, Yue, Qin, Ping, and Hu operasw-chinese-opera-qupai: fixed-tune systems used in forms such as Yuan drama, Ming-Qing chuanqi, Kunqu, and Sichuan opera's high-pitched style
The series layer includes concrete timing references, including broadcast page anchors around 17–18, 30, 45, and 60 pages, as well as minute-marked pilot breakdowns. It also covers the practical difference between a story that works once and a series engine capable of generating many more episodes.
3. Tradition and industry
The third layer places technique in cultural and commercial context:
- American feature case studies
- Japanese screenwriting methods
- Korean and French writing approaches
- Chinese television production practice
- Industry, pitching, options, credits, agents, WGA context, and buyer-facing development
This layer is especially useful when an agent needs to compare methods rather than present one supposedly universal formula. The repository preserves disagreements between sources, such as Douglas's four-act grid and Oberg's argument that act breaks are primarily determined by story pressure and production realities.
4. Master corpora
The final layer treats complete works as working material rather than inspirational examples. It includes:
- Chekhov's full-length plays and one-acts
- Six Ozu screenplays
- All four seasons of Succession, covering 39 shooting scripts
- Case studies from The West Wing, The Sopranos, Downton Abbey, Fleabag, and other series
The succession-series-writing skill, for example, examines pressure chambers, ceremony-based container episodes, downward humiliation, stage directions carrying subtext, reversals triggered by a single word, alternative lines, and the challenge of engineering an ending after the main engine loses pressure.
Practical usage examples
A writer can use the library with natural-language requests:
Break my 12-episode series into acts and place the act-outs.
This should route toward sw-series-structure.
My dialogue is all on the nose. Diagnose the problem and rewrite the scene.
This combines sw-dialogue with sw-scene-craft.
Turn this 400,000-word novel into a 40-episode Chinese drama outline.
This calls for sw-chinese-series-practice and sw-series-engine-bible, with adaptation decisions made explicit rather than silently compressing the source.
Set up a project and keep track of where I am.
This is the role of sw-workflow, which uses a story-bible convention to preserve project state across sessions.
Why the multilingual design uses one source tree
The skill bodies are written in Chinese because much of the source material consists of Chinese originals or Chinese translations. However, the agent's response language follows the writer's request. A writer can ask in English, Japanese, Korean, or French and receive an answer in that language.
This avoids maintaining separate translated skill trees. The project previously had an English edition, but maintaining parallel trees creates a long-term synchronization problem: once one translation exists, users reasonably request Japanese, Korean, French, Russian, and more. Five language trees would mean hundreds of files that can drift independently.
Terminology is handled through mappings in sw-workflow/terms.md. Concepts such as “logline,” “act out,” “beat sheet,” “showrunner,” and “staff writer” remain anchored to their original industry terms. Terms without a clean equivalent can retain the original word with a gloss.
There is a trade-off: readers who cannot read Chinese cannot directly audit the underlying instruction files. The translated READMEs provide installation, structure, and policy information, while the full skill tables and bibliography remain in the main and Chinese documentation.
Detailed reference material without oversized skill files
Each skill contains a SKILL.md with principles, workflows, and checklists. Almost every skill also includes a reference.md containing tables, examples, excerpts, and worked analyses.
Some larger areas are split across multiple reference files. For example, the series case studies divide material into English script collections, pilot breakdowns, and Asian texts. The series-engine skill separates engine teardowns, document field tables, and completed samples.
This separation keeps the operational instructions manageable. The repository also aims to keep SKILL.md files under roughly 40 KB and skill descriptions under the 1,024-character limit required by the project conventions.
Adding new media without weakening the core
The repository uses a strict rule for deciding whether a new genre deserves its own skill: it must change the structure, format, or language rules of writing. A subject category such as spy thriller or family drama does not automatically qualify. Those belong in case studies unless a complete, distinctive corpus justifies a dedicated treatment.
A new medium should add:
- One new skill
- A boundary table explaining which general methods transfer
- One entry in the workflow's routing table
The general skills should not be rewritten every time a new medium is added. This architecture allowed the series layer and Chinese opera layer to expand without changing the basic premise, structure, dialogue, and scene skills.
Chinese opera illustrates why this separation matters. Banqiang and qupai are treated as two systems because their vocal structures, lyric methods, and stage units differ. Regional traditions within one system can often use reference appendices rather than separate skills, preventing hundreds of near-duplicate skills for every regional genre.
Source base and licensing
The project draws from 17 screenwriting craft books, 15 television-writing books, published screenplays and plays, and extensive Chinese opera theory, scripts, and scores. Sources include work by Syd Field, Blake Snyder, Robert McKee, Lajos Egri, Lisa Cron, Neil Landau, Pamela Douglas, William Rabkin, Jesse Armstrong, Aaron Sorkin, Phoebe Waller-Bridge, Chekhov, Ozu, and many Chinese writers and theorists.
The repository is licensed for personal study use. Quotations remain the property of their original authors and translators, so it is best treated as a learning and development tool rather than a replacement for licensed source texts.
Who should use it?
This collection is a strong fit for:
- Screenwriters developing features or pilots
- Television writers building a series bible
- Showrunners and writers-room participants
- Researchers comparing dramatic traditions
- Developers testing domain-specific agent skills
- Writers adapting novels, plays, or existing intellectual property
- Educators building structured writing exercises
It is less suited to users looking for a one-click “write me a script” prompt. Its value comes from staged development, explicit craft decisions, source-backed alternatives, and repeated revision.
With 1,000 stars and 118 forks listed on GitHub, the repository has attracted substantial interest as an example of how specialized knowledge can be packaged for coding agents. Its most important idea is architectural: instead of asking a general model to remember every screenwriting rule at once, give it small, composable skills that can be selected according to the current writing problem.