Research Project Init

Research Project Init Skill

Toolchain

Python

Rust

Mixed projects (Python + Rust)

Both toolchains are active. Python bindings to Rust are handled via maturin and PyO3. The Python package wraps the Rust crate; uv manages the Python environment.

Bootstrap forked or existing repos

When the human opens a repo that is already populated (forked, cloned, or otherwise pre-existing) but lacks the Claude Code infrastructure, skip the full scaffold and run only the bootstrap procedure below. This procedure must be run at session start on any forked repo that does not already have .claude/hooks/pre-commit-gate.sh.

B1 — Detect what is missing

Check each of the following. Only create what does not already exist.

.claude/
.claude/settings.json
.claude/hooks/pre-commit-gate.sh
.claude/agents/implementer.md
.claude/agents/test-writer.md
.claude/agents/physics-reviewer.md   (if physics project)
CLAUDE.md

B2 — Create .claude/ and settings.json

mkdir -p .claude

Write .claude/settings.json using the template from the hooks skill. Invoke the hooks skill to do this — do not write it from memory.

B3 — Write the pre-commit gate hook

Write .claude/hooks/pre-commit-gate.sh using the template from the hooks skill. Make it executable: chmod +x .claude/hooks/pre-commit-gate.sh.

The hook template in the hooks skill is the canonical source. It includes:

• Python checks (ruff, ty, pytest) if pyproject.toml exists
• Rust checks (cargo fmt --check, cargo clippy -- -D warnings, cargo test) if Cargo.toml exists
• notes/ guard — blocks commits of private working notes
• Branch guard — blocks commits directly to main

B4 — Install agent definitions

Invoke the agent-team skill to write agent definition files into .claude/agents/. This step is language-agnostic: the agent-team skill detects the project language from the repo contents.

B5 — Generate or verify CLAUDE.md

Invoke the project-md skill to generate CLAUDE.md. If one already exists, verify it covers the required sections (project type, physics context, toolchain, conventions) and offer to update it.

B6 — Verify the gate

Run the gate checks appropriate for the project language and confirm they pass before proceeding to any implementation work:

Rust: cargo fmt --check && cargo clippy -- -D warnings && cargo test Python: uv run ruff check . && uv run ruff format --check . && uv run ty check && uv run pytest

Commit the infrastructure:

git add .claude/ CLAUDE.md
git commit -m "chore: bootstrap Claude Code infrastructure on forked repo"
git push

Step-by-step init procedure (new projects)

1 — Gather inputs

Ask the human for:

1. Project language — Python, Rust, or both?
2. Project name — this becomes the repo name and the primary package/crate name (use underscores for Python packages, hyphens for Rust crates and repos, e.g. repo quantum-sensing, Python package quantum_sensing, Rust crate quantum-sensing).
3. One-sentence description — used in pyproject.toml/Cargo.toml and README.md.
4. W&B? — Python and mixed projects only: "Will this project use Weights & Biases?" (yes/no).
   • If yes: "What is your W&B entity?" and "What is your W&B project name?"

Do not proceed until these are confirmed.

2 — Create directory structure

Adapt based on LANGUAGE. Create only the directories applicable to the language.

Python projects (LANGUAGE = python)

<repo-name>/
├── <package_name>/          # Python package (named after the project)
│   └── __init__.py
├── scripts/                 # Standalone entrypoint scripts (not part of package)
│   └── .gitkeep
├── data/                    # Data files — tracked via Git LFS
│   └── .gitkeep
├── plots/                   # Output figures — gitignored
│   └── .gitkeep
├── notes/                   # Private notes — gitignored
│   └── .gitkeep
├── tests/
│   └── __init__.py
├── configs/                 # Config files (YAML, TOML)
│   └── .gitkeep
├── .github/
│   ├── workflows/
│   │   └── ci.yml
│   └── PULL_REQUEST_TEMPLATE.md
├── .claude/
│   ├── agents/
│   ├── hooks/
│   └── settings.json
├── .gitignore
├── .gitattributes           # Git LFS tracking rules
├── .pre-commit-config.yaml
├── pyproject.toml
├── README.md
└── .env.example             # Always; .env itself is gitignored

Rust projects (LANGUAGE = rust)

<repo-name>/
├── src/
│   └── lib.rs               # or main.rs for a binary crate
├── tests/                   # Rust integration tests
│   └── .gitkeep
├── data/                    # Data files — tracked via Git LFS
│   └── .gitkeep
├── plots/                   # Output figures — gitignored
│   └── .gitkeep
├── notes/                   # Private notes — gitignored
│   └── .gitkeep
├── configs/                 # Config files (YAML, TOML)
│   └── .gitkeep
├── .github/
│   ├── workflows/
│   │   └── ci.yml
│   └── PULL_REQUEST_TEMPLATE.md
├── .claude/
│   ├── agents/
│   ├── hooks/
│   └── settings.json
├── .gitignore
├── .gitattributes           # Git LFS tracking rules
├── Cargo.toml
└── README.md

Note: no scripts/, no <package_name>/, no .env.example for pure Rust projects unless W&B or other secrets are in use (unlikely for Rust-only).

Mixed projects (LANGUAGE = both)

<repo-name>/
├── <package_name>/          # Python package
│   └── __init__.py
├── src/                     # Rust crate (e.g. PyO3 extension module)
│   └── lib.rs
├── scripts/                 # Python entrypoint scripts
│   └── .gitkeep
├── data/
│   └── .gitkeep
├── plots/
│   └── .gitkeep
├── notes/
│   └── .gitkeep
├── tests/
│   └── __init__.py          # Python tests; Rust unit tests are inline in src/
├── configs/
│   └── .gitkeep
├── .github/
│   ├── workflows/
│   │   └── ci.yml
│   └── PULL_REQUEST_TEMPLATE.md
├── .claude/
│   ├── agents/
│   ├── hooks/
│   └── settings.json
├── .gitignore
├── .gitattributes
├── .pre-commit-config.yaml
├── pyproject.toml
├── Cargo.toml
├── README.md
└── .env.example

3 — Initialise uv and pyproject.toml

uv init <repo-name>
cd <repo-name>

pyproject.toml:

[project]
name = "<repo-name>"
version = "0.1.0"
description = "<one-sentence description>"
requires-python = ">=3.14"
dependencies = []

[build-system]
requires = ["hatchling"]
build-backend = "hatchling.build"

[tool.hatch.build.targets.wheel]
packages = ["<package_name>"]

[tool.ruff]
line-length = 88
target-version = "py314"

[tool.ruff.lint]
select = ["E", "F", "I", "UP", "B", "SIM"]
ignore = []

[tool.ruff.format]
quote-style = "double"

[tool.ty]
# ty configuration — add per-project overrides here

[tool.pytest.ini_options]
testpaths = ["tests"]

Add dev dependencies:

uv add --dev ruff ty pytest pre-commit

4 — .gitignore

Emit all sections applicable to the project language.

# Environment
.env

# Output (regenerable)
plots/
notes/

# Editors
.vscode/
.idea/
*.swp

# OS
.DS_Store
Thumbs.db

# ── Python (include if LANGUAGE = python or both) ──────────────────────────────
.venv/
__pycache__/
*.py[cod]
.uv/

# W&B (if applicable)
wandb/

# ── Rust (include if LANGUAGE = rust or both) ──────────────────────────────────
target/
# Note: commit Cargo.lock for binaries; do not commit it for libraries.
# Ask the human which applies, and gitignore Cargo.lock only for libraries.

5 — Git LFS

git lfs install

.gitattributes:

*.npy    filter=lfs diff=lfs merge=lfs -text
*.npz    filter=lfs diff=lfs merge=lfs -text
*.hdf5   filter=lfs diff=lfs merge=lfs -text
*.h5     filter=lfs diff=lfs merge=lfs -text
*.pt     filter=lfs diff=lfs merge=lfs -text
*.pkl    filter=lfs diff=lfs merge=lfs -text
*.csv    filter=lfs diff=lfs merge=lfs -text

Note: data/ is not gitignored. Files within it are committed via LFS. This ensures reproducibility. Warn the human if LFS is not available on their remote.

6 — pre-commit

Python projects (LANGUAGE = python or both)

.pre-commit-config.yaml:

repos:
  - repo: https://github.com/astral-sh/ruff-pre-commit
    rev: v0.4.4
    hooks:
      - id: ruff
        args: [--fix]
      - id: ruff-format

  - repo: https://github.com/astral-sh/ty
    rev: 0.0.0-alpha.6    # update to latest stable
    hooks:
      - id: ty

Rust projects (LANGUAGE = rust or both)

Rust does not use pre-commit for cargo fmt and cargo clippy — these are handled by the Claude Code pre-commit gate hook (see hooks skill) and the GitHub Actions CI workflow. This is because pre-commit's Rust support requires the system Rust toolchain rather than the project-pinned version, which can produce inconsistent results.

For mixed projects, use the Python .pre-commit-config.yaml above for ruff/ty, and rely on the Claude Code hook and CI for Rust checks.

Install hooks (Python and mixed projects only):

uv run pre-commit install

7 — GitHub Actions CI

Python projects — .github/workflows/ci.yml

name: CI

on:
  push:
    branches: ["main"]
  pull_request:
    branches: ["main"]

jobs:
  check:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
        with:
          lfs: true

      - name: Install uv
        uses: astral-sh/setup-uv@v4

      - name: Install dependencies
        run: uv sync --dev

      - name: Lint (ruff)
        run: uv run ruff check .

      - name: Format check (ruff)
        run: uv run ruff format --check .

      - name: Type check (ty)
        run: uv run ty check

      - name: Tests (pytest)
        run: uv run pytest

Rust projects — .github/workflows/ci.yml

name: CI

on:
  push:
    branches: ["main"]
  pull_request:
    branches: ["main"]

env:
  CARGO_TERM_COLOR: always

jobs:
  check:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
        with:
          lfs: true

      - name: Install Rust stable
        uses: dtolnay/rust-toolchain@stable
        with:
          components: rustfmt, clippy

      - name: Cache cargo
        uses: Swatinem/rust-cache@v2

      - name: Format check
        run: cargo fmt --check

      - name: Clippy
        run: cargo clippy -- -D warnings

      - name: Tests
        run: cargo test

Mixed projects — .github/workflows/ci.yml

name: CI

on:
  push:
    branches: ["main"]
  pull_request:
    branches: ["main"]

env:
  CARGO_TERM_COLOR: always

jobs:
  rust:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
        with:
          lfs: true

      - name: Install Rust stable
        uses: dtolnay/rust-toolchain@stable
        with:
          components: rustfmt, clippy

      - name: Cache cargo
        uses: Swatinem/rust-cache@v2

      - name: Format check
        run: cargo fmt --check

      - name: Clippy
        run: cargo clippy -- -D warnings

      - name: Tests
        run: cargo test

  python:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
        with:
          lfs: true

      - name: Install uv
        uses: astral-sh/setup-uv@v4

      - name: Install dependencies
        run: uv sync --dev

      - name: Lint (ruff)
        run: uv run ruff check .

      - name: Format check (ruff)
        run: uv run ruff format --check .

      - name: Type check (ty)
        run: uv run ty check

      - name: Tests (pytest)
        run: uv run pytest

8 — PR template

.github/PULL_REQUEST_TEMPLATE.md:

## Why
<!-- What problem does this solve? What decision was made and why? -->

## How
<!-- What approach was taken? Key design decisions. -->

## What
<!-- Concrete list of changes: files added/modified, functions changed, behaviour altered. -->

## Tests
<!-- What is tested? How? If no tests: justify why. -->

## Physics & numerical considerations
<!-- Discretisation choices, stability, units, boundary conditions, known limitations.
     Write "N/A" only if the code has zero physical or numerical content. -->

## Checklist
- [ ] Types pass (`ty check`)
- [ ] Lint passes (`ruff check`)
- [ ] Tests pass (`pytest`)
- [ ] Data files not committed accidentally (only via LFS)
- [ ] No hardcoded paths or secrets

9 — W&B (if yes)

Create .env:

WANDB_API_KEY=
WANDB_ENTITY=<entity>
WANDB_PROJECT=<project-name>

Create .env.example (committed to repo):

WANDB_API_KEY=
WANDB_ENTITY=<entity>
WANDB_PROJECT=<project-name>

Create configs/wandb.yaml:

entity: <entity>
project: <project-name>

Create <package_name>/logging.py:

"""Experiment logging utilities."""

import os
from pathlib import Path

import wandb
from dotenv import load_dotenv

load_dotenv()


def init_run(config: dict, name: str | None = None) -> wandb.sdk.wandb_run.Run:
    """Initialise a W&B run with the given config."""
    return wandb.init(
        entity=os.environ["WANDB_ENTITY"],
        project=os.environ["WANDB_PROJECT"],
        name=name,
        config=config,
    )

Add W&B and python-dotenv as dependencies:

uv add wandb python-dotenv

Add wandb/ to .gitignore (already included in template above).

10 — README.md

# <repo-name>

<one-sentence description>

## Installation

Requires [uv](https://github.com/astral-sh/uv).

```bash
git clone https://github.com/<owner>/<repo-name>
cd <repo-name>
uv sync

Development

uv run pytest          # run tests
uv run ruff check .    # lint
uv run ty check        # type check

Pre-commit hooks are installed automatically on first git commit after uv sync.

Structure

---

### 11 — Initialise git and push

All repositories are created under the `QuantumSensing` organisation. The remote
URL is always:

```bash
git init
git add .
git commit -m "chore: initialise project scaffold"
git branch -M main
git remote add origin https://github.com/QuantumSensing/<repo-name>.git
git push -u origin main

Remind the human to create the GitHub repo under QuantumSensing before pushing if it does not yet exist: https://github.com/organizations/QuantumSensing/repositories/new

12 — Invoke project-md skill

After the repo is pushed, invoke the project-md skill to generate CLAUDE.md. This requires answers to the project type and context questions defined in that skill. Commit the result:

git add CLAUDE.md
git commit -m "chore: add CLAUDE.md project context"
git push

13 — Invoke agent-team skill

Based on the PROJECT_TYPE declared during CLAUDE.md generation, invoke the agent-team skill to install agent definition files into .claude/agents/. Commit:

git add .claude/agents/
git commit -m "chore: add Claude Code agent definitions"
git push

14 — Invoke hooks skill

Invoke the hooks skill to write .claude/settings.json and .claude/hooks/pre-commit-gate.sh. Commit:

git add .claude/
git commit -m "chore: add Claude Code pre-commit gate hook"
git push

15 — Invoke github-projects skill

Hand off to the github-projects skill to complete project setup:

1. Delete default GitHub labels and create: Writing, Coding, Experiments, Research, Won't Fix.
2. Create a GitHub Project named after the repository under QuantumSensing.
3. Configure the board with columns: Backlog → Todo → In Progress → In Review → Done.
4. Link the repository to the project.

Follow the Project initialisation section of the github-projects skill.

Post-init checklist (present to human)

Project scaffold complete. Before starting work:

[ ] Create the GitHub repo at https://github.com/organizations/QuantumSensing/repositories/new
[ ] Run: git push -u origin main
[ ] Add WANDB_API_KEY to .env (if using W&B)
[ ] Add GITHUB_TOKEN to .env — must have repo + project scopes
[ ] Confirm Git LFS is enabled on the remote (Settings → Git LFS)
[ ] Install pre-commit hooks: uv run pre-commit install
[ ] Confirm GitHub Project board columns are set correctly (or set manually if API unavailable)
[ ] Board URL: https://github.com/orgs/QuantumSensing/projects/{project_number}
[ ] Review CLAUDE.md and update physics/ML context if needed