new-project

name	new-project
description	Initialize a new research project by copying the template, then run an interactive wizard to fill PROJECT.md with conversation language, topic, research questions, paper/thesis type, target venue, citation style, output language, search languages, word budget, and deadline.
disable-model-invocation	true
argument-hint	[type] [subdir]

Initialize a New Project

This skill creates a new project directory and guides you through the initial scoping: topic definition, research questions, output format selection, and word budget allocation.

PROCEDURE

1. Check for Existing Project (GATE)

Check: Does PROJECT.md exist in current directory or in [subdir]?

• If YES and it has real content (not all TBD): ABORT. This fires before Field 0 (Conversation Language) is ever asked, so print bilingually:

  ❌ Bu dizinde zaten doldurulmuş bir PROJECT.md var. Yeni bir proje için:
  1. Başka bir dizine geçin veya
  2. /new-project subdir-adi komutuyla yeni alt dizin oluşturun veya
  3. Mevcut PROJECT.md dosyasını silip yeniden başlayın.
  
  ❌ A filled-in PROJECT.md already exists in this directory. For a new project:
  1. Switch to another directory, or
  2. Create a new subdirectory with /new-project <subdir-name>, or
  3. Delete the existing PROJECT.md and start over.
  

• If NO or [subdir] does not exist: proceed.

2. Copy Template — projects are SELF-CONTAINED

A project must work as its own Claude Code workspace, so it needs the kit runtime, not just the data skeleton. Locate the kit root (the directory containing this skill: walk up from .claude/skills/new-project/ — call it $KIT).

1. Target: <subdir>/ if provided (mkdir -p <subdir>), else the current directory. The target must NOT be inside $KIT/project-template.
2. Copy INTO the target:
   • $KIT/project-template/* (PROJECT.md, sources/, evidence/, analysis/, bib/, outline/, drafts/, reviews/, exports/, build/, and the non-manuscript axes: proposal/, teaching/, dossier/)
   • $KIT/CLAUDE.md (constitution — sessions in the project dir load it)
   • $KIT/.claude/ (agents, skills, hooks, settings.json)
   • $KIT/scripts/ (all python scripts/... calls resolve locally)
   • $KIT/templates/ (CSL styles, reference.docx, thesis skeletons)
3. Do NOT copy: $KIT/data/ (large index datasets — instead, if $KIT/data/*.csv exists, create a small pointer so venue_check finds it: on any OS run python -c "import pathlib; pathlib.Path('data').mkdir(exist_ok=True)" then copy the CSVs OR tell the user to re-run bash INSTALL.sh --update-data inside the project. Prefer copying if disk allows; datasets are per-machine, not per-project.)
4. If target == current directory AND it already IS the kit root (CLAUDE.md + scripts/ present), skip copying runtime files — only copy project-template/*.
5. Patch the hook interpreter in the copied settings.json. The copied .claude/settings.json wires every hook to a literal interpreter name (python). On a machine where only python3 or py resolves, every hook fails to launch — and PreToolUse hooks fail OPEN, so enforcement is silently absent rather than noisily broken. bash INSTALL.sh does this for a cloned kit; a project created from a plugin install never runs INSTALL, so do it here. From the target directory, detect the working interpreter and run:

   python scripts/patch_hook_interpreter.py --target <python|python3|py>
   

   Use whichever of python / python3 / py actually launched the command. Skip only when step 4 applied (the kit root is already patched).
6. Confirm: "Project initialized at <path> (self-contained: runtime + templates copied, hooks wired to <interpreter>). Open Claude Code IN this directory for all project work."

3. Run Interactive Wizard

Walk the user through each field of PROJECT.md (present each as a numbered choice or text entry). Ask Field 0 first, before anything else — it decides which language the rest of this wizard (and every later skill's gates/aborts/summaries) is conducted in.

Field 0: Conversation Language

Hangi dilde konuşalım? / Which language should we work in?
1) Türkçe (varsayılan / default)
2) English

Seçiminiz / Your choice: <input>
→ Write "tr" or "en" to the `- **Conversation language:**` bullet.
→ From THIS POINT ON — the rest of this wizard, and every other skill's
  user-facing text for the life of the project — conduct in the chosen
  language. Every other skill's SKILL.md writes its gate/abort/summary
  messages in Turkish as the kit's default; when Conversation language is
  "en", present the equivalent content in natural English (translate
  meaning and structure — headings, bullet shape, ✅/❌ markers — rather
  than reproducing the Turkish text verbatim). This is orthogonal to
  Output language (Field 6 below): Conversation language governs how the
  assistant talks to the user; Output language governs the language the
  manuscript itself is drafted in — a thesis can be drafted in English
  while the working conversation stays in Turkish, or vice versa.

Field 1: Topic (Turkish or English)

Konunuz nedir? (örnek: "CNN tabanlı rulman arıza teşhisi")
→ Validate: not empty, not single word. If unclear: ask for 1–2 more sentences of context.
→ Write to PROJECT.md's Identity section, `- **Title (working):**` bullet.

Field 2: Research Questions (≥1, ≤4)

Araştırma sorularınız nelerdir? (satır satır girin, sonra boş satır ile tamamlayın)
RQ1: <input>
RQ2: <input>
...
→ Validate: each ≥10 words, ends with "?". If fewer than 1 or >4: ask to reconsider.
→ Write: one `  - RQ<n>: <text>` bullet per RQ under "Research questions:" —
  this exact bullet shape (leading `-`, `RQ<n>:` prefix) is what
  word_budget.py --rq-trace's regex matches; do not reformat it.

Field 3: Paper/Thesis Type

Çıktı türü:
1) makale (IMRaD) / article (empirical)
2) sistematik derleme / systematic review
3) derleme (tematik) / narrative review
4) scoping derleme / scoping review
5) niteliksel çalışma / qualitative study
6) tez (BSc/MSc/PhD)
7) konferans bildirisi / conference paper
8) diğer (hangi?)

Seçiminiz: <input>
→ Validate against paper-types skill list. If invalid: show list again.
→ Write to the `- **Type:**` bullet (keep the `**bold**` markers — word_budget.py's
  regex matches on them). Use the paper-types enum key, not the numbered label
  (e.g. "empirical-imrad", not "makale (IMRaD)").

Field 4: Target Venue

Hedef dergi/konferans/kurum (adı veya ISSN; isteğe bağlı):
<input>
→ If provided: suggest running `/journal-check "<input>"` after wizard.
→ Write to the `- **Target venue:**` bullet.

Field 5: Citation Style

Atıf stili:
1) APA 7
2) IEEE
3) Chicago (Author-Date)
4) Vancouver/AMA
5) Harvard
6) MLA 9
7) özel CSL (provide CSL file later)

Seçiminiz: <input>
→ Validate against citation-styles skill list.
→ Write to the `- **Citation style:**` bullet using the enum key (apa7 | mla9 |
  chicago-author-date | chicago-notes | ieee | vancouver | harvard | csl:<file>) —
  the same key citation-styles' CSL filename map (Section "CSL File Mapping")
  resolves at render time.

Field 6: Output Language

Akademik çıktının dili:
1) English (İngilizce)
2) Türkçe
3) Her ikisi (Özet + Abstract)

Seçiminiz: <input>
→ Write to the `- **Output language:**` bullet as "tr", "en", or "both".

Field 7: Search Languages (for literature review)

Literatür taraması hangi dillerde yapılsın?
1) İngilizce (EN)
2) Türkçe (TR)
3) Her ikisi (EN + TR) [önerilir]

Seçiminiz: <input>
→ Explain: "EN tabanlı konular: her ikisi önerilir. Türkiye-spesifik konular: TR zorunlu."
→ Write to the `- **Search languages:**` bullet as "tr", "en", or "both". This
  sets the default the bilingual decision matrix (literature-review skill)
  starts from — literature-scout may still override per-topic and must log why.

Field 8: Word Budget (approximate total)

Yaklaşık hedef sözcük sayısı (örnek: 6500, 40000, vb.):
<input>
→ Validate: numeric, 1000–200000. If outside: suggest reasonable range for type.
→ Write to the `- **Word budget (total):**` bullet as a plain integer (no
  thousands separator inside the value word_budget.py's regex captures — e.g.
  "6500", not "6.500" — the script strips commas/dots for TR-locale input but
  writing the clean digit string avoids relying on that).

Field 9: Deadline (optional)

Teslim tarihi (YYYY-MM-DD) [isteğe bağlı]:
<input>
→ Validate: if provided, must be future date. If past: warn and ask to confirm.
→ Write to the `- **Deadline:**` bullet (or leave "TBD" if blank).

4. Summary & Confirmation

Display a summary table (in the Field-0 conversation language):

PROJECT SUMMARY
───────────────────────────────────────
Conversation Lang:  [tr|en]
Topic:              [user topic]
RQs:                [count]
Type:               [type]
Target Venue:       [venue or "TBD"]
Citation Style:     [style]
Output Language:    [lang]
Search Languages:   [langs]
Word Budget:        [count] words
Deadline:           [date or "No deadline"]
───────────────────────────────────────

Ask: "Doğru mu? (evet/hayır)"

• evet: proceed to step 5.
• hayır: ask which field to re-enter and loop back to that field.

5. Write PROJECT.md

project-template/PROJECT.md was already copied into the project root in Step 2. Edit that file in place (Edit tool, not a rewrite) — fill each TBD with the wizard's answer, keeping the section headings, HTML comments, and - **Label:** bullet shapes byte-for-byte as shipped in the template.

Do NOT rewrite it as YAML. word_budget.py (--rq-trace and the plain budget report) parses this file with regexes anchored to the Markdown bold-label format — Word budget \(total\):\*{0,2}, \*\*Type:\*\*, and ^\s*[-*]\s*(RQ\d+)\s*: for the RQ bullets. A YAML version matches none of these: every RQ trace and word-budget check would silently return empty/None for the life of the project.

Field → template-bullet mapping (all under existing section headings — do not add new ones):

conversation_lang -> ## Languages         - **Conversation language:**
topic            -> ## Identity          - **Title (working):**
type             -> ## Identity          - **Type:**
RQs              -> ## Identity          - **Research questions:** (one `  - RQ<n>: <text>` bullet each)
target_venue     -> ## Target & format   - **Target venue:**
citation_style   -> ## Target & format   - **Citation style:**
output_language  -> ## Languages         - **Output language:**
search_languages -> ## Languages         - **Search languages:**
word_budget      -> ## Budget & schedule - **Word budget (total):**
deadline         -> ## Budget & schedule - **Deadline:**

Leave - **Per-section budgets:** auto, - **Template:** none, - **Own data:** none, - **Compute notes:** TBD, and - **Phase:** scoping exactly as shipped — later skills/scripts update them, the wizard does not. Append a trailing HTML comment (invisible to the regex parsers, but useful for humans) recording creation time:

<!-- created_at: <ISO timestamp> -->

6. Final Message

Print to user, in the Field-0 conversation language (Turkish shown; if "en" was chosen, translate meaning/structure rather than reproducing this text):

✅ Proje başlatıldı! 

Sonraki adımlar:
1. Hedef dergisi için: /journal-check "<venue>"
2. Literatür taraması başlatmak için: /lit-review [--systematic]
3. Proje durumu: /status

Açılış PROJECT.md: <path>/PROJECT.md

---

GATES SUMMARY

Gate	Trigger	Action
Existing filled PROJECT.md	Yes	ABORT with Turkish explanation; offer alternatives
Topic empty/too short	Yes	REJECT, ask for clarification
RQ count invalid	<1 or >4	REJECT, explain reasoning
RQ quality	Single word / <10 words	WARN, suggest clarification
Type invalid	Not in paper-types list	ABORT, show options
Style invalid	Not in citation-styles list	ABORT, show options
Lang invalid	Not EN/TR/both	ABORT, show options
Word budget invalid	<1000 or >200000	WARN, suggest range for type
Deadline invalid	Past date	WARN, ask to confirm

---

ERROR MESSAGES (Turkish default — translate to English when Field 0 is "en")

❌ Topic başlığınız çok kısa. Lütfen en az bir cümle yazın.
❌ RQ sayısı 1–4 arasında olmalı. Kaç tane RQ'nuz var?
❌ Atıf stili tanınmıyor. /citation-styles skill'ine bakarak desteklenen stilleri görün.
❌ Bu dizinde zaten doldurulmuş PROJECT.md var.

---

SCRIPT USAGE

None. This is a pure wizard. No Python scripts run in this step.

---

FOLLOWING STEPS

After PROJECT.md is created:

• User may run /journal-check to verify target venue quality.
• User may run /lit-review to start literature search.