---
description: Import an existing manuscript (docx, markdown, txt, or directory) and structure it into a Scriven .manuscript/ directory.
argument-hint: "<file_or_directory_path> [--type <work_type>]"
---

# Import

You are importing an existing manuscript into Scriven. The writer has content -- in a Google Doc, a Scrivener project, a folder of markdown files, a Word document, or similar -- and wants Scriven's tooling without starting from scratch.

## What to do

1. **Locate the file or directory.** Accept: .docx, .md, .txt, .rtf, a directory of any of these.

2. **Read the content.** For docx, use pandoc or similar to extract text. For markdown, read directly. For a directory, concatenate files in order.

3. **Detect the work type.** If `--type` was passed, use it. Otherwise analyze the content and make a best guess:
   - Scene headers (`INT./EXT.`) -> screenplay
   - Chapter markers with prose -> novel or memoir
   - Section headers with citations -> academic paper
   - Verse and chapter numbers -> scripture
   - Panel descriptions -> comic
   - Poem titles separated by breaks -> poetry collection
   - Ask the writer to confirm.

4. **Parse structure.** Identify the unit boundaries based on the work type's `command_unit` from CONSTRAINTS.json. For a novel, find chapter breaks. For a screenplay, find act breaks or scene headers. For a research paper, find section headers.

5. **Run voice analysis.** Take a 2000-word sample from the imported text and extract the writer's voice DNA: sentence architecture, vocabulary register, figurative density, POV, tense, dialogue style. Populate STYLE-GUIDE.md with the detected values. Flag anything uncertain for the writer to confirm.

6. **Detect characters.** Scan for proper nouns that appear as speakers in dialogue or as recurring agents in narration. Build a draft CHARACTERS.md with name, estimated role (protagonist, antagonist, supporting), and detected voice patterns. Flag for writer review.

7. **Generate the .manuscript/ directory.** Create all context files (WORK.md, BRIEF.md, OUTLINE.md, RECORD.md, STYLE-GUIDE.md, CHARACTERS.md, etc.) populated from the import. Also copy `WRITING-RULES.md` verbatim from the installed Scriven templates (`templates/WRITING-RULES.md`) into `.manuscript/` so the drafter, voice-checker, and originality-check have the canonical universal rules available. Save the actual drafted text as `.manuscript/drafts/body/{N}-{A}-DRAFT.md` files, one per atomic unit.

   Populate `RECORD.md` from the imported manuscript, not from speculation. Extract established on-page facts, open threads, promises, continuity facts, reader expectations, and visible character, subject, argument, procedure, image, object, or relationship movement. Mark uncertain discoveries as "needs writer confirmation" instead of treating them as settled truth.

8. **Set STATE.md** to reflect that all imported units are drafted but not yet reviewed. This lets the writer pick up with `/scr:editor-review` or `/scr:next`.

9. **Generate a report.** Tell the writer:
   ```
   Imported [filename] as a [work_type].

   Structure detected: X chapters, Y scenes total
   Word count: 65,432
   Voice profile: extracted from 2000-word sample (review in STYLE-GUIDE.md)
   Characters detected: 7 (review in CHARACTERS.md)
   Record initialized: established facts and open threads extracted into RECORD.md

   Some things I wasn't sure about:
   - [specific flags: uncertain chapter breaks, possible POV shifts, etc.]

   Next steps:
   - Review STYLE-GUIDE.md and CHARACTERS.md -- confirm or adjust
   - Run /scr:next to start working on the imported manuscript
   ```

## Edge cases

- **Messy source.** If the import has weird formatting, OCR errors, or inconsistent structure, do your best and flag the issues. Don't refuse the import.
- **Huge manuscript.** For texts over 200k words, warn about import time but proceed. Split voice analysis across multiple samples.
- **Mixed work types.** A Google Doc with both prose and outline notes -- import as primary work type, save notes to `.manuscript/notes/`.
- **Source in a different language.** Detect it. Ask if the writer wants to work in the source language or translate first.
- **No clear structure.** If you can't find chapter or scene breaks, ask the writer how they want the text divided before creating draft files.

## What NOT to do

- Don't "improve" the writer's text during import. Preserve it exactly. They'll revise via normal commands.
- Don't rename files without asking.
- Don't skip voice analysis -- it's the highest-value part of import.
- Don't assume character roles without evidence. Mark uncertain roles as such.

## Response Contract

Every writer-facing response must end with one to four next-command suggestions. Each suggestion must include a short explanation of what that path will do.

Use this format:

```markdown
Next commands:
- `/scr:...`: One short sentence explaining what this path will do.
- `/scr:...`: One short sentence explaining what this alternate path will do.
```

If exactly one path is clearly best, provide one suggestion. If two, three, or four useful paths exist, show them as alternatives. Do not force a linear path when the writer has a real choice.

If the writer seems unsure or no specific next command is obvious, include this default option:

```markdown
Next commands:
- `/scr:next`: Inspect the project state and choose the right next step.
```

If the command stops because a prerequisite is missing, suggest the command that fixes the prerequisite. Keep every explanation practical and writer-facing.

## Tone

Respectful. The writer has worked on this text. You're bringing it into a new system, not judging it. Report what you found clearly, flag your uncertainties honestly, and make clear the writer is in charge of confirming.