`/new-project`¶

Scaffolds a new empirical project folder with LaTeX, code, scripts, results, literature and guidance subfolders plus a README and a project-level CLAUDE.md.

Pack: ai-asset-pricing (Alex Dickerson)

Category: infra

Field: finance

Updated: 2026-04-19

Stages: —

↗ view SKILL.md on source · GitHub stars

Create New Research Project¶

Four phases: Validate → Gather Context → Create Scaffold → Setup Paper.

Examples¶

/new-project momentum_replication -- create project with name only
/new-project vol_surface "Implied volatility surface dynamics" -- name + description
/new-project bond_liquidity -- another example

Phase 1: Validate Input¶

Parse $ARGUMENTS: - First token = project name - Remaining tokens (if any) = optional description text

If $ARGUMENTS is empty, use AskUserQuestion to ask:

"What should the project be called? Use lowercase with underscores (e.g., momentum_replication, vol_surface, bond_liquidity)."

Validation rules: - Must match regex: ^[a-z][a-z0-9_]{1,49}$ (lowercase, underscores only, starts with letter, 2-50 chars) - Reject Windows-reserved names: con, prn, aux, nul, com1-com9, lpt1-lpt9 - If invalid, explain the rules, show examples, and re-ask

Collision check: - If projects/<name>/ already exists, ask the user: "Project <name> already exists. Do you want to open it, or choose a different name?" - Do NOT overwrite existing projects

Phase 2: Gather Context¶

If no description was provided in $ARGUMENTS, use a single AskUserQuestion with these questions: 1. "Brief description of the project (1-2 sentences):" 2. "Which WRDS databases will this project use? (e.g., CRSP, OptionMetrics, Compustat, TAQ, Bonds, JKP)" 3. "Any initial methodology notes or reminders for this project?"

If a description WAS provided in $ARGUMENTS, still ask questions 2 and 3.

Store the answers for template filling.

Phase 3: Create Scaffold¶

3a. Create directories¶

Run a single Bash command:

Bash

mkdir -p projects/<name>/{latex,code,scripts/tests,results/{figures,tables},literature,guidance,_misc}

3b. Write project README.md¶

Use the Write tool to create projects/<name>/README.md:

Markdown

## {Project Name in Title Case}

{User's description}

### Status

- [ ] Data sourced
- [ ] Exploratory analysis complete
- [ ] Main results produced
- [ ] Write-up drafted
- [ ] Write-up finalized

### Data Dependencies

This project uses datasets from the global `data/` folder:

| Dataset | Description | Status |
|---------|-------------|--------|
| *(to be filled as data is fetched)* | | |

### Structure

| Folder | Purpose |
|--------|---------|
| `latex/` | LaTeX writeup |
| `code/` | Production code (clean, reusable) |
| `scripts/tests/` | Exploratory investigations (each in own subfolder with `output/`) |
| `results/` | Publication-ready figures and tables |
| `literature/` | Reference papers |
| `guidance/` | Methodology notes |
| `_misc/` | Catch-all |

### Notes

- Created: {today's date, YYYY-MM-DD}
- WRDS databases: {databases}

3c. Write project CLAUDE.md¶

Use the Write tool to create projects/<name>/CLAUDE.md:

Markdown

### Project: {project_name}

{User's description}

### Data Sources

WRDS databases: {databases}

Data is pulled from the global `data/` folder. Datasets used by this project:
- *(list grows as data is fetched)*

### Conventions

#### Scripts and Tests
- Exploratory work goes in `scripts/tests/{test_name}/`
- Each test gets its own subfolder with an `output/` dir for all artifacts
- Name test folders descriptively: `summary_stats`, `return_predictability`, `cross_section_reg`, etc.

#### Output Formats
- **Figures:** PNG (300 DPI) for drafts, PDF for final versions in `results/figures/`
- **Figure generation:** Use `docs/ai/figures.md` and `fintools.figures`; use
  `style="fins"` for house-style publication plots and `style="ft"` for
  FT-style output
- **Tables:** LaTeX `.tex` for the paper, CSV for debugging
- **Final outputs** for the paper go in `results/figures/` and `results/tables/`

#### Code
- `code/` = clean, reusable production code
- `scripts/` = one-off investigations and scratch work
- Import from `code/` into `scripts/`, never the reverse

#### LaTeX
- Main document: `latex/main.tex`
- Reference results: `../results/figures/`, `../results/tables/`
- BibTeX: `latex/references.bib`

### Paper-Specific Terminology

*(Define paper-specific preferred/banned terms here. Example:)*
| Use | Instead of | Reason |
|-----|------------|--------|
| *(e.g., "execution risk")* | *(e.g., "trading friction")* | *(our umbrella term)* |

### LaTeX Section Keys

*(Register `%% BEGIN/END` section marker keys here for `/latex-doctor` verification.)*
| Key | Section |
|-----|---------|
| `introduction` | Introduction |

### Target Journal

*(e.g., JF, RFS, JFE — used by `/submission-prep` for journal-specific checks.)*

### Key Claims to Verify at Submission

*(List the main quantitative claims that `/submission-prep` should verify against tables/text.)*
- *(e.g., "58 of 341 factors are significant at 5%")*

### Project-Specific Instructions

{User's notes, or "None yet -- add methodology reminders, conventions, or constraints here."}

3d. Write project .gitignore¶

Use the Write tool to create projects/<name>/.gitignore:

Text Only

## LaTeX build artifacts
latex/*.aux
latex/*.bbl
latex/*.blg
latex/*.log
latex/*.out
latex/*.fdb_latexmk
latex/*.fls
latex/*.synctex.gz
latex/*.toc

## Python
__pycache__/
*.pyc
.ipynb_checkpoints/

## R
.Rhistory
.RData

## OS
.DS_Store
Thumbs.db

3e. Print scaffold summary¶

Print the directory tree to the user:

Text Only

### Project Created: {project_name}

projects/{project_name}/
  README.md
  CLAUDE.md
  .gitignore
  latex/
  code/
  scripts/
    tests/
  results/
    figures/
    tables/
  literature/
  guidance/
  _misc/

Phase 4: Setup Paper¶

After printing the scaffold summary, ask the user:

"Set up the LaTeX paper now? This copies the boilerplate template into latex/ with your title and [REMOVE]-tagged exemplar content."

Use AskUserQuestion with options: - Yes — proceed to set up the paper - No — skip; user can run /setup-paper later

If Yes: 1. Derive a default paper title from the project name (replace underscores with spaces, title case) 2. cd into projects/<name>/ so that /setup-paper detects it as the active project 3. Invoke /setup-paper with the derived title — it will auto-target latex/ 4. The user can provide --authors and --topic at this point if they want

If No, print:

Text Only

Next steps:
  1. Run /setup-paper "Your Title" to scaffold the LaTeX paper when ready
  2. Add literature PDFs to literature/
  3. Start an investigation: create scripts/tests/{test_name}/ with an output/ subfolder
  4. Fetch data using WRDS agents — it saves to data/ and you reference it from here
  5. Add methodology notes to guidance/

/new-project¶

Create New Research Project¶

Examples¶

Phase 1: Validate Input¶

Phase 2: Gather Context¶

Phase 3: Create Scaffold¶

3a. Create directories¶

3b. Write project README.md¶

3c. Write project CLAUDE.md¶

3d. Write project .gitignore¶

3e. Print scaffold summary¶

Phase 4: Setup Paper¶

`/new-project`¶