Setup

Project Setup

> Run this skill once at the start of your dissertation project. > It checks prerequisites, asks configuration questions, and generates the full project structure.

Step 1: Check Zotero MCP Connection

Before anything else, verify that Zotero MCP is connected.

Test: Attempt to call zotero_search_items (or equivalent) with a trivial query like "test".

If the call succeeds:

• Report: "✓ Zotero MCP is connected."
• Proceed to Step 1b.

If the call fails or the tool is not available:

Step 1a: Auto-add Zotero MCP to Claude Code

Attempt to register Zotero MCP automatically by running:

claude mcp add zotero -- uvx zotero-mcp serve

If this command succeeds, report:

Zotero MCP has been added to your Claude Code configuration.
Please restart Claude Code and run /setup again for the connection to take effect.

Then STOP. The user must restart Claude Code for the new MCP to load.

If the auto-add command fails (e.g., uvx not found, network error):

• Report:

``` ✗ Could not auto-add Zotero MCP.

This scaffold requires Zotero MCP for literature management. Please install it manually:

OPTION 1 — If you have uv installed: Run in your terminal (outside Claude Code): claude mcp add zotero -- uvx zotero-mcp serve

OPTION 2 — If you installed via pip: claude mcp add zotero -- zotero-mcp serve

OPTION 3 — Install from scratch: pip install zotero-mcp-server claude mcp add zotero -- zotero-mcp serve

For other Zotero MCP options see: • zotero-mcp (recommended): https://github.com/54yyyu/zotero-mcp • zotero-mcp plugin: https://github.com/cookjohn/zotero-mcp • mcp-zotero: https://github.com/kaliaboi/mcp-zotero

ALSO RECOMMENDED — Desktop Commander MCP: https://github.com/wonderwhy-er/DesktopCommanderMCP Install: npx @wonderwhy-er/desktop-commander@latest setup

Desktop Commander gives Claude full filesystem access and native PDF reading capabilities, improving reliability of reading Zotero PDFs.

After installing, restart Claude Code and run /setup again. ```

• Ask: "Would you like to continue setup without Zotero, or stop and install it first?"
• If continuing without Zotero: note in CLAUDE.md that Zotero is not connected, and the /missing-lit skill will use web search only. Skip Steps 1b and 2, proceed to Step 3.

Step 1b: Check Better BibTeX

Only if Zotero MCP is connected. Better BibTeX (BBT) is required for citekey-based paper lookup.

Test: Take any item returned from the Step 1 search. Call zotero_get_item_metadata with format: bibtex on that item. Check whether the bibtex output begins with @type{citekey, (e.g., @article{becker2019,).

If the bibtex output contains a citekey:

• Report: "✓ Better BibTeX is installed."
• Proceed to Step 2.

If the bibtex output has no citekey, returns an error, or looks malformed:

• Report:

``` ✗ Better BibTeX does not appear to be installed in Zotero.

This scaffold requires Better BibTeX (BBT) citekeys for paper lookup. Without BBT, Claude cannot reliably identify papers in your library.

Install Better BibTeX: https://retorque.re/zotero-better-bibtex/

Steps:

1. Download the .xpi file from the link above
2. In Zotero: Tools → Add-ons → Install Add-on From File → select the .xpi
3. Restart Zotero
4. BBT will auto-generate citekeys for all items in your library

After installing, restart Claude Code and run /setup again. ```

• Ask: "Would you like to continue setup without Better BibTeX, or stop and install it first?"
• If continuing without BBT: note the limitation in CLAUDE.md. Literature note creation will require manual citekey entry rather than auto-lookup.

Step 2: Configure Zotero PDF Access

Only if Zotero MCP is connected. Ask the user:

> What is the full path to your Zotero PDF storage folder? > This is where Zotero stores PDF attachments on your machine. > Examples: > - macOS with Google Drive: /Users/yourname/Library/CloudStorage/GoogleDrive-you@gmail.com/My Drive/Zotero/Library > - macOS local: /Users/yourname/Zotero/storage > - Linux: /home/yourname/Zotero/storage > - Windows: C:\Users\yourname\Zotero\storage

Save the answer to literature/zotero-config.md:

---
pdf_folder: "{user's answer}"
bbt_verified: {true|false}
date_configured: {YYYY-MM-DD}
---

# Zotero Configuration

- **pdf_folder**: Absolute path to Zotero PDF storage directory.
  Claude reads PDFs from this location during literature note creation.
  Update this path if your Zotero library moves.
- **bbt_verified**: Whether Better BibTeX was detected during setup.

Then report:

✓ Zotero configured.

Literature notes will be stored in literature/zotero-notes/ as @citekey.md files.
When you say "analyse paper @citekey", Claude will:
  1. Find the paper in Zotero via MCP
  2. Access the PDF from: {pdf_folder}
  3. Split it into 4-page chunks and read in batches
  4. Classify the paper and extract structured information
  5. Write the note to literature/zotero-notes/@citekey.md

When working on a specific paper (e.g., Paper 1), Claude will copy
relevant notes from literature/zotero-notes/ to paper-1/literature/zotero-notes/.

Optional: If you open this project folder in Obsidian, all markdown notes
are browsable with graph view. Install LaTeX Suite for equation rendering.

Step 3: Gather Project Configuration

Ask the user the following questions. Wait for answers before proceeding.

1. What is your dissertation topic? (2–3 sentences)
2. How many publishable chapters/papers will your dissertation include?

(Common: 3 for economics, but could be 2, 4, or more depending on your program)

1. For each paper, provide:

• A short title or topic description
• The likely identification strategy (DiD, IV, RDD, SC, structural, descriptive, or "undecided")

1. What is your primary analysis language? (R / Stata / Python)
2. Your name, institution, and supervisor's name
3. Target journal tier (Top 5, top field, or strong field — or specific journal names)
4. Your research field and subfields (for the domain profile)

Step 4: Generate Paper Directories

For each paper (based on the number provided in Step 3):

1. Copy the entire paper-template/ directory to paper-N/ (where N is 1, 2, 3, ...)
2. Update paper-N/CLAUDE.md with the paper-specific topic, strategy, and phase
3. Update paper-N/pre-analysis-plan/strategy-memo.md header with the paper's question
4. Update paper-N/code/R/00_master.R (or .do / .py) header with the paper number
5. Update paper-N/quality_reports/research_journal.md with the paper number
6. Update paper-N/replication-package/README.md with the paper number

Also update dissertation/chapters/ to include one chapter stub per paper:

• paper-1.tex, paper-2.tex, ..., paper-N.tex
• Update dissertation/main.tex to \input{} all paper chapters

Step 5: Configure Master Files

1. CLAUDE.md — Fill in all placeholders:

• Project title, institution, supervisor
• All paper topics and research questions
• Current phase (typically "SETUP" at this point)

1. MEMORY.md — Leave template structure, no changes needed yet

1. .claude/references/domain-profile.md — Fill in:

• Field and subfields
• Common identification strategies
• Known data sources (ask the user if unsure)
• Target journals
• Seminal references (ask the user for 5–10 foundational papers)

1. .claude/references/journal-profiles.md — Review pre-populated profiles, ask if any need adjustment for the user's specific subfield

1. .claude/references/notation-registry.md — Start with core variables if the user already has conventions; otherwise leave as template

Step 6: Update Dissertation Structure

Based on the paper count:

1. Update dissertation/main.tex to include the correct number of \input{chapters/paper-N} lines
2. Create chapter stubs for each paper in dissertation/chapters/
3. Update dissertation/cross-paper-checks.md column headers for the actual number of papers

Step 7: Clean Up Template

After all paper directories are generated:

1. Delete the paper-template/ directory (it's no longer needed — each paper-N/ is independent)
2. Or rename it to _paper-template/ if the user wants to keep it for future reference

Step 8: Initial Commit

git add -A
git commit -m "scaffold: project setup — [N] papers on [topic]"

Step 9: Report

Present a summary:

• Number of papers created
• Zotero connection status
• Obsidian connection status
• Files configured
• What to do next (suggested first task: process literature, or draft a strategy memo)

Important

• Follow all file safety rules during setup
• Do not modify the template files — copy them, then modify the copies
• Ask for user confirmation before committing

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: saptarsibhowmick
• Source: saptarsibhowmick/claude-workflow-for-econ-phd
• License: MIT

Install and usage instructions live in the source repository linked above.