paper-spine-intake

name: paper-spine-intake description: Collects PaperSpine workflow options and writes config for flash/pro, scene, language, and inputs.

PaperSpine Intake

Use this skill before substantive PaperSpine work when the workflow is not yet configured.

Required Output

Create:

• paper_rewriting_output/paper_spine_config.json
• paper_rewriting_output/paper_spine_config.md

The JSON must contain:

{
  "workflow": "rewrite_existing",
  "scene": "journal",
  "tier": "flash",
  "output_language": "en",
  "target_name": "",
  "materials_dir": "",
  "draft_path": "",
  "user_motivation": "",
  "official_urls": [],
  "special_requirements": [],
  "word_output": "none",
  "translation_package": "none",
  "reference_mode": "local_first",
  "reference_paths": ["."],
  "citation_target_count": 20
}

Allowed values:

• workflow: rewrite_existing, build_from_materials
• scene: journal, conference, report_review, competition
• tier: flash, pro
• output_language: en, zh
• word_output: none, docx
• translation_package: none, zh; use zh only when output_language is en
• reference_mode: local_first, specified_paths, web
• reference_paths: local folders/files to index for references; default .
• citation_target_count: final intended citation count; default 20

Interaction Rules

The command-line wizard is the supported UI for Claude Code and Codex CLI. When a real terminal is available, route through paper-spine-ui and use the bundled external terminal wizard; do not downgrade to asking the user to hand-write JSON or type every option in chat.

In Claude Code, do not run python scripts/intake_wizard.py directly from an agent Bash/tool call. That execution surface may not expose stdin and can hang. Use /paperspine for normal work; it launches this intake UI automatically when configuration is missing. The UI is a real terminal TUI with Up/Down option switching, Left/Right field switching, Enter edit/confirm, and S save. Claude Code does not currently expose a public third-party skill API for embedding this custom keyboard UI inside the chat input box itself. If needed, run the launcher relative to this skill directory:

powershell.exe -NoProfile -ExecutionPolicy Bypass -File scripts/launch_paperspine_ui.ps1 -OutputDir paper_rewriting_output

This opens a separate interactive PowerShell window. For a user-run terminal, this direct command is also supported:

python scripts/intake_wizard.py --output-dir paper_rewriting_output

The wizard provides arrow-key menus in a real terminal, falls back to numbered menus when needed, shows descriptions for each option, runs a final review screen and edit loop, and writes both JSON and Markdown config.

Use native structured questions only if the host exposes them reliably in the current session. If neither terminal execution nor native questions are available, then and only then ask the user to answer with the fallback template.

Read references/interactive-intake.md for the exact question order and fallback template.

Defaults

• journal and conference default to en.
• Chinese course, Chinese competition, and explicit Chinese writing requests default to zh.
• If the user has a draft, choose rewrite_existing.
• If the user has a folder of experiment settings, results, figures, notes, PDFs, Word files, or reports, choose build_from_materials.
• If the user does not choose tier, default to flash.
• If the user does not request Word output, default word_output to none.
• If the paper is English, offer translation_package: zh as an optional Chinese reading package for intermediate artifacts and final structure.
• Default reference_mode to local_first.
• Default reference_paths to ["."] unless the user provides paths.
• Default citation_target_count to 20.

Global UI Language

On first setup, or when the user asks to change PaperSpine interface language, run the wizard in global setup mode:

python scripts/intake_wizard.py --setup-global --output-dir paper_rewriting_output

Use the wizard's terminal UI for all config fields. Prefer the arrow-key UI in a real terminal; use numbered command-line menus as fallback. Avoid asking the user to manually type JSON/options unless terminal execution is impossible.