JSPM

  • ESM via JSPM
  • ES Module Entrypoint
  • Export Map
  • Keywords
  • License
  • Repository URL
  • TypeScript Types
  • README
  • Created
  • Published
  • Downloads 244
  • Score
    100M100P100Q90687F
  • License MIT

Spec-driven creative writing, publishing, and translation pipeline for AI coding agents. From blank page to published book.

Package Exports

    This package does not declare an exports field, so the exports above have been automatically detected and optimized by JSPM instead. If any package subpath is missing, it is recommended to post an issue to the original package (scriven-cli) to support the "exports" field. If that is not possible, create a JSPM override to customize the exports field for this package.

    Readme

    Scriven

    Spec-driven creative writing, publishing, and translation for AI coding agents.

    I don't outline -- Claude does. I don't edit -- Claude does. I don't format -- Claude does. I just write.

    Inspired by GSD (get-shit-done), Scriven brings spec-driven workflows to creative writing. It runs inside your AI coding agent (Claude Code, Cursor, Gemini CLI, and more) and handles everything from blank page to publication-ready package.

    npx scriven@latest

    What is this

    Scriven is a command system that turns your AI coding agent into a full creative writing studio. It supports 46 work types -- novels, screenplays, research papers, poetry collections, scripture commentaries, comics, memoirs -- each with its own adaptive vocabulary and toolset.

    You get 101 slash commands covering every stage of the writing pipeline:

    • Create -- Set up a project with tailored context files. Progressive onboarding, never overwhelming.
    • Write -- Discuss, plan, draft, and revise one unit at a time. The drafter agent loads your Voice DNA and writes in your voice, not generic AI prose.
    • Polish -- Editor review, line edit, copy edit, continuity check, voice check, beta reader simulation, sensitivity review.
    • Publish -- Front/back matter, cover art, blurbs, query letters, KDP packages, IngramSpark packages, EPUB, PDF, Fountain, Final Draft, LaTeX.
    • Translate -- Deep translation with glossary management, cultural adaptation, back-translation verification, multi-language simultaneous publishing.
    • Collaborate -- Parallel revision tracks, co-writing workflows, continuity merge checking.

    Everything adapts to your work type. A novel uses /scr:draft for chapters. A screenplay uses /scr:draft for acts. A Quran commentary uses /scr:draft for surahs. Same command, tradition-native vocabulary.


    Quick start

    # Install
    npx scriven@latest
    
    # In any project directory, open Claude Code (or your preferred agent) and:
    /scr:new-work        # Start a fresh project
    /scr:demo            # Explore a pre-built sample first
    /scr:next            # The universal "what should I do now" command
    /scr:help            # See what's available for your work type

    If you only ever type /scr:next, you can complete an entire novel. It always knows what's next.


    The Voice DNA system

    Scriven's core insight: drafted prose should sound like you, not like an AI. Before drafting begins, /scr:profile-writer builds a detailed voice profile across 15+ dimensions:

    • Narrative perspective, tense, narrator stance
    • Sentence architecture, paragraph rhythm
    • Vocabulary register, figurative density, recurring image systems
    • Dialogue style, character voice differentiation
    • Pacing, transitions, emotional range
    • Do/don't/consider rules specific to the writer

    This profile is saved as STYLE-GUIDE.md and loaded into every drafter agent invocation. The drafter writes one atomic unit per fresh context -- a scene, a subsection, a passage -- with the style guide as its primary reference. Voice stays consistent across hundreds of scenes.

    For sacred and historical texts, Voice DNA is supplemented by 10 sacred voice registers (prophetic, wisdom, legal, liturgical, narrative-historical, apocalyptic, epistolary, psalmic, parabolic, didactic).


    Work types supported

    Prose: novel, novella, short story, flash fiction, memoir, creative nonfiction, biography, essay, essay collection

    Script: screenplay, stage play, TV pilot, TV series bible, audio drama, libretto/musical

    Academic: research paper, thesis/dissertation, journal article, white paper, literature review, monograph

    Visual: comic, graphic novel, children's book, picture book

    Poetry: poetry collection, single poem, song/lyric

    Interactive: interactive fiction, game narrative

    Sacred & historical: scripture (Biblical, Quranic, Torah, Vedic, Buddhist, generic), commentary/exegesis, devotional, liturgical text, historical chronicle, historical account, mythological collection, religious epic, sermon, homiletic collection

    Each work type has its own structural hierarchy. Commands rename automatically -- a Torah commentary uses /scr:plan-parashah, not /scr:plan-chapter.


    Autopilot mode

    Run the full pipeline autonomously. Three profiles:

    • Guided -- Pause after each unit for review
    • Supervised -- Batch through several units, pause for review
    • Full-auto -- Run until complete; only pause on critical failures
    /scr:autopilot --profile supervised

    Plus:

    • /scr:autopilot-publish -- One command generates front matter, back matter, cover, and full export package
    • /scr:autopilot-translate french german spanish -- Simultaneous multi-language editions

    Writer mode vs. developer mode

    Scriven detects non-technical writers and hides git terminology. Instead of commit, branch, merge, diff -- you see save, version, compare, accept changes. All the power, none of the coding jargon.

    Technical writers can enable developer mode in settings for full git access and verbose output.


    Philosophy

    Scriven is built on five principles:

    1. The writer's voice is sacred. The drafter never imposes generic AI style. Every drafted sentence passes through the Voice DNA gate.

    2. Fresh context per atomic unit. Each scene, subsection, or passage is drafted in a clean context. This prevents voice drift, context bloat, and keeps each unit at its best.

    3. Progressive disclosure. Onboarding asks 3 questions, not 30. Depth is optional and always additive.

    4. Tradition-native vocabulary. A Quran commentary uses surahs and ayahs. A Bible study uses books and verses. A screenplay uses acts and scenes. The tool adapts to the tradition -- the writer never adapts to the tool.

    5. /scr:next always works. The universal interface. A writer who only ever types /scr:next can complete an entire novel, from blank page to KDP package.


    Documentation


    Supported AI coding agents

    • Claude Code (primary, most complete)
    • Cursor (full support)
    • Gemini CLI (full support)
    • Codex CLI (full support)
    • OpenCode (full support)
    • GitHub Copilot (full support)
    • Windsurf (full support)
    • Antigravity (full support)
    • Manus Desktop (full support)

    Status

    Version: 1.0.0

    All features shipped -- 101 commands across 46 work types, with voice profiling, autopilot drafting, full publishing pipeline, illustration generation, translation, collaboration tools, and sacred text support. Runs on 9 AI coding agent runtimes.


    License

    MIT. See LICENSE.

    Contributing

    Scriven is an open project. Contributions welcome -- especially new work types, additional runtime adapters, and voice register definitions for languages and traditions we haven't covered yet.