Superpaper Init

Once answers are collected, scaffold the full project structure:

<project>/
  CLAUDE.md
  AGENTS.md
  GEMINI.md

  meta/
    memory.md          # Initialize with project creation entry
    progress.md        # All chapters marked as [NOT STARTED]
    milestones.md      # Empty template
    plan.md            # Empty template
    style/
    tasks/

  writings/
    ch01-<name>/
      outline.md
      resources.md
      draft.md
      status.md
      assets/
      reviews/
    ch02-<name>/
      ...

  references/
    literature/
    results/
    docs/

  .gitignore

File generation steps:

1. Create all directories. Use ch01-<slug>, ch02-<slug>, ... based on chapter names provided.

2. Generate meta/memory.md:

# Memory

## Current Focus
Project just initialized: <title>

## Recent Decisions
- <date>: Project created with <citation style> citation style, language: <language>

## Open Questions

## Key Insights

## Next Session
- Run `checkin` to begin your first session
- Consider running `brainstorm` to explore research directions

3. Generate meta/progress.md:

# Progress

| Chapter | Status | Notes |
|---------|--------|-------|
| ch01 — <name> | [NOT STARTED] | |
| ch02 — <name> | [NOT STARTED] | |
...

4. Generate meta/milestones.md:

# Milestones

| Milestone | Target Date | Status |
|-----------|-------------|--------|
| Topic finalized | | [ ] |
| Literature review complete | | [ ] |
| Outline approved | | [ ] |
| First draft complete | | [ ] |
| Revisions complete | | [ ] |
| Submission | | [ ] |

5. Generate meta/plan.md as an empty template with headers: # Plan, ## Timeline, ## Weekly Routine.

6. Generate writings/ch0N-<name>/status.md for each chapter:

# <Chapter Name> Status

## Sections
(to be filled during outline)

## Blockers

## Last Updated
<date> by init

6b. Generate writings/ch0N-<name>/outline.md for each chapter as an empty placeholder:

# Chapter N: <Chapter Name>

## Outline

*To be drafted. Run `superpaper-outline` to plan this chapter.*

6c. Generate writings/ch0N-<name>/draft.md for each chapter as an empty placeholder:

# Chapter N: <Chapter Name>

## Draft

*Not yet started. Complete the outline first, then run `superpaper-draft`.*

6d. Generate writings/ch0N-<name>/resources.md for each chapter as an empty placeholder:

# Chapter N: <Chapter Name> — Resources

## Literature

*No references yet. Run `superpaper-ingest` to add literature.*

## Results

*No analysis results yet.*

7. Generate CLAUDE.md with full project settings (see template below).

8. Generate AGENTS.md referencing the same project settings and skill mapping as CLAUDE.md.

9. Generate GEMINI.md with tool mapping reference (activate_skill maps to Skill tool in Claude Code).

10. Generate .gitignore (see template below).

11. Run:

# If git not already initialized, run git init first
git init   # safe to run even if already initialized (no-op)
git add .
git commit -m "init: <title>"

Note: Check if .git/ already exists. If so, skip git init to avoid confusing output — run git add . && git commit -m "init: <title>" directly.

12. Print next steps:

Project scaffolded. Run checkin to start your first session, or brainstorm to explore research directions.

Mode 2: Adopt Existing Repo (init adopt)

Key principle: adopt mode never moves or deletes existing files without explicit confirmation from the user. This is non-negotiable — NEVER alter or remove existing content without asking.

For repos with existing content:

1. Paper language? (zh / en)

2. Paper title?

3. Citation style? (GB/T 7714 / APA / Chicago / IEEE / custom)

4. Scan the repo for content patterns:

   • Markdown files → potential chapter drafts
   • PDF files → potential literature copies
   • Code/data directories → potential results/analysis
   • Zotero exports (CSL-JSON, BibTeX .bib files) → literature references
   • Existing directory structure

5. Present proposed mapping to the user. Example:

Found content:
  chapters/intro.md  →  writings/ch01-introduction/draft.md (symlink/record in CLAUDE.md)
  chapters/lit.md    →  writings/ch02-literature/draft.md
  data/              →  references/results/
  refs.bib           →  references/literature/ (after ingest)

No files will be moved. Mapping recorded in CLAUDE.md.
Confirm? (yes / adjust)

6. After user confirmation, create the meta/ layer on top:

   • meta/memory.md — initial entry describing the adoption and what was found
   • meta/progress.md — chapters with existing drafts marked as [IN PROGRESS], others as [NOT STARTED]
   • Per-chapter status.md for each identified chapter

7. If the existing structure uses different directory names (e.g., chapters/ instead of writings/), record the mapping in CLAUDE.md under a ## Directory Mapping section. Do not rename directories.

8. Generate CLAUDE.md / AGENTS.md / GEMINI.md.

9. Commit: "init: adopt existing repo — <title>"

CLAUDE.md Template

Generate this in the user's project root:

# <Paper Title>

## Project Settings
- Language: <language>
- Citation style: <citation style>
- University: <from question 5, or blank if user skipped>

## Skills
This project uses superpaper.
Before any task, check for a matching skill.

### Project Manager
- Session start → superpaper-checkin
- New materials → superpaper-ingest
- End of session → superpaper-memory-update

### Research Facilitator
- Topic exploration → superpaper-brainstorm
- Literature gaps → superpaper-research
- Data questions → superpaper-data-explore / superpaper-analysis

### Writing Assistant
- Chapter planning → superpaper-outline
- Writing → superpaper-draft
- Reviewing → superpaper-review

### Quality Assurance
- Audit → superpaper-lint
- Submission prep → superpaper-export

## Context Loading
1. Read meta/memory.md — recent history and decisions
2. Read meta/progress.md — overall state
3. Read relevant writings/<ch>/status.md — chapter details
4. Read relevant references/<key>/summary.md — don't re-read originals

## Writing Principles
- Mark uncertainty: [TODO], [CITE], [DATA]. Never fabricate.
- Always structurally complete. Missing sections have placeholders.
- Respect human edits. Never overwrite without asking.
- Follow meta/style/style-guide.md — extracted from sample paragraphs/chapters you provide.
- Shorter is better.
- Task files in `meta/tasks/` should have a status header: `status: pending | in-progress | done`. Stale done tasks can be archived to `meta/tasks/archive/`.

## Boundaries
- Do not fabricate data, results, or citations
- Do not mark chapters as finalized (human only)
- Do not make major structural changes without asking
- Do not delete human content without permission
- For deep research: generate prompts, don't pretend to search
- For coding/analysis: generate specs, don't write code

.gitignore Template

# PDFs (markdown-first — PDFs are local reference copies)
*.pdf

# Data files
*.csv
*.xlsx
*.dta
*.sav
*.rds

# Large binary files
*.zip
*.tar.gz

# Assets (figures regenerated from code)
writings/*/assets/*
!writings/*/assets/.gitkeep

# OS
.DS_Store
Thumbs.db

# Editor
.idea/
.vscode/
*.swp

# Exported drafts (regenerate with export skill)
export/

Inputs

• User's answers to interactive prompts (language, title, citation style, chapter structure)
• For adopt mode: existing repo content

Outputs

• Full project directory structure
• meta/memory.md, meta/progress.md, meta/milestones.md, meta/plan.md
• Per-chapter writings/ch0N-<name>/ directories with placeholder files
• references/literature/, references/results/, references/docs/
• CLAUDE.md, AGENTS.md, GEMINI.md with project settings and skill mapping
• .gitignore
• Initial git commit

Notes

• Chapter directories: ch01-<slug>, ch02-<slug>, etc. Slugify names (lowercase, hyphens).
• Generated files should match the project language setting (use Chinese headings for zh projects).
• The generated CLAUDE.md is what any agent reads when entering the dissertation project.
• For adopt mode: if no title is found, ask before proceeding.
• Suggest checkin as the very next skill to run after init completes.