Claude Code Template

Template for senior Python projects with Claude Code configured deterministically. Implements the three-layer system: advisory (CLAUDE.md + rules), deterministic (hooks), and on-demand (skills + commands + agents).

Create a new project from this template

Quick way (1 command):

gh repo create my-new-project --template DrLux/claude-code-template --private --clone
cd my-new-project

From GitHub: click "Use this template" on github.com/DrLux/claude-code-template, then clone the created repo.

Both methods create a repo with a clean history, without template commits.

After creation:

# 1. Customize CLAUDE.md with your project's stack and invariants
# (replace all placeholders [PROJECT NAME], [e.g. FastAPI], etc.)

# 2. Initialize the project context
touch .claudememory       # will be populated by /memory-update
touch CLAUDE.local.md     # personal notes and PR feedback — never commit

# 3. First session: generate the architectural map
/architect

Once per machine (global developer identity):

cp templates/global-memory.md ~/.claude/memory.md
# Edit ~/.claude/memory.md to match your profile

How to use this template (manual setup)

# 1. Clone the template into the new project directory
git clone https://github.com/DrLux/claude-code-template.git my-project
cd my-project

# 2. Remove the template history and initialize your repo
rm -rf .git
git init

# 3. Customize CLAUDE.md with your project's stack and invariants
# (replace all placeholders [PROJECT NAME], [e.g. FastAPI], etc.)

# 4. Copy the global developer identity (once per machine)
cp templates/global-memory.md ~/.claude/memory.md
# Edit ~/.claude/memory.md to match your profile

# 5. Initialize the project context
touch .claudememory       # will be populated by /memory-update
touch CLAUDE.local.md     # personal notes and PR feedback — never commit

# 6. First session: generate the architectural map
/architect

Structure

.
├── .claude/
│   ├── settings.json           — hook configuration
│   ├── hooks/                  — deterministic scripts (linting, tests, security)
│   ├── skills/                 — commands with side effects (commit)
│   ├── commands/               — on-demand commands (architect, pythonic, docstrings)
│   ├── rules/                  — thematic rules loaded automatically
│   └── agents/                 — specialized subprocesses (reviewer, test-writer)
├── templates/
│   └── global-memory.md        — template for ~/.claude/memory.md (global, do not commit)
├── CLAUDE.md                   — project manifest (committed, shared with team)
├── CLAUDE.local.md             — personal notes (gitignored)
├── ARCHITECT.md                — architectural map (generated by /architect)
└── .claudememory               — dynamic project state (gitignored)

Control plane

Layer	Files	Behavior
Advisory	CLAUDE.md, .claude/rules/	Claude reads and interprets
Deterministic	.claude/hooks/	Executed automatically, non-bypassable
On-demand	.claude/skills/, .claude/commands/	Activated only on explicit invocation

Available commands

Skill/Command	Function
/commit	Assisted commit with human confirmation (never pushes)
/architect	Rescans the repo and regenerates ARCHITECT.md (auto-triggers on ExitPlanMode)
/pythonic	Idiomatic Python 3.10+ refactoring
/docstrings	Adds Google-style docstrings to public APIs
/memory-update	Updates .claudememory with the session state

Active hooks

Hook	Trigger	Action
SessionStart	Session start	Loads .claudememory + git log
PreToolUse	Every Bash command	Blocks destructive patterns
PostToolUse	.py file write	ruff --fix + mypy
PostToolBatch	End of write batch	cross-file mypy
ExitPlanMode	Plan approval	Auto-runs /architect to update architecture map
Stop	Session end	pytest + ARCHITECT.md update

Global developer identity setup

Copy templates/global-memory.md to ~/.claude/memory.md and customize:

• Seniority Contract: adapt to your level and domain
• Technical Philosophy: reflects your preferences
• Toolchain Standard: modify if you use different stacks

Last updated: 01/06/2026