Personal Claude Code Dev Setup

A personal, user-level configuration that applies to every project on your machine. One setup. All projects. Full software development lifecycle — from idea to merge.

---

Install

Windows (PowerShell):

.\install.ps1

Mac/Linux:

bash install.sh

---

The full workflow

PLAN ──→ SPEC ──→ DIVIDE ──→ START/BUILD ──→ UPDATE (if needed) ──→ DONE ──→ CLOSE ──→ REVIEW ──→ (you merge)
 │         │         │          │              │                  │        │        │
 ↓         ↓         ↓          ↓              ↓                  ↓        ↓        ↓
README   SPEC.md  OVERVIEW  feature       append-only        handoff   docs +   PR ready
 .md              + .tasks/  branches      task/spec sync               cleanup

Step	Command	Agent	Output
Plan the project/feature	/plan	planner	README.md
Write the technical spec	/spec	spec	SPEC.md
Divide work into tasks	/divide [name]	divider	Manager loop + .tasks/OVERVIEW.md + per-agent task files
Implement (per agent)	/start role=X branch=Y	backend/frontend/etc	Code on feature branch
Sync mid-task change	/update [reason]	current agent	Append-only spec/task revision
Finish a task	/done	current agent	Handoff summary
Hand off to next agent	/switch role=X	—	Handoff note + new role loaded
Finalize feature state	/close [name]	current agent	Main README/SPEC alignment + task cleanup decision
Review before merge	/review	current agent	PR description + security/QA findings
Update style preferences	/style [feedback]	—	Updated style file

Notes:

• /divide now runs a pre-gate manager loop (all involved roles raise questions/pitfalls before task files are finalized).
• On Windows, this setup assumes bash is available on PATH (Git Bash recommended) for hook execution.

---

All 10 agents

Agent	Model	Purpose
planner	Opus 4.6	Produces README.md. Interviews you, never writes code
spec	Opus 4.6	Produces SPEC.md from the plan. Never writes code
divider	Sonnet 4.6	Produces .tasks/ from plan + spec. Never writes code
backend	Sonnet 4.6	APIs, services, middleware, auth, integrations
database	Sonnet 4.6	Schema, migrations, queries, indexes
frontend	Sonnet 4.6	Components, pages, styling, state, forms
devops	Sonnet 4.6	CI/CD, Docker, Kubernetes, infra, secrets
security	Opus 4.6	Read-only security review, severity-ranked findings
qa	Haiku 4.5	Tests, coverage, regression testing
researcher	Opus 4.6	Live docs via Context7, persistent memory

---

All 10 commands

Command	Usage	What it does
/plan	/plan or /plan new feature: X	Starts planning session → README.md
/spec	/spec or /spec [feature]	Starts spec session → SPEC.md
/divide	/divide [feature-name]	Divides work → .tasks/ files
/start	/start role=backend branch=auth/backend	Begins implementation session
/update	/update backend needs new column	Mid-task append-only updates to task/spec context
/done	/done	Self-review checklist + handoff summary
/switch	/switch role=database	Clean role handoff
/close	/close auth	Finalize main README, SPEC alignment, and task cleanup
/review	/review	Pre-merge security + QA + PR description
/style	/style [feedback]	Update code style preferences

---

File structure

~/.claude/                           ← installed by install.sh
├── CLAUDE.md                        ← master rules, loaded every session
├── settings.json                    ← plan mode, safety deny rules, hooks
├── .claude.json (home dir)          ← MCP servers
│
├── agents/
│   ├── planner.md                   ← Opus 4.6
│   ├── spec.md                      ← Opus 4.6
│   ├── divider.md                   ← Sonnet 4.6
│   ├── backend.md                   ← Sonnet 4.6
│   ├── database.md                  ← Sonnet 4.6
│   ├── frontend.md                  ← Sonnet 4.6
│   ├── devops.md                    ← Sonnet 4.6
│   ├── security.md                  ← Opus 4.6
│   ├── qa.md                        ← Haiku 4.5
│   └── researcher.md                ← Opus 4.6 + persistent memory
│
├── commands/
│   ├── plan.md, spec.md, divide.md  ← lifecycle commands
│   ├── start.md, update.md, done.md ← session commands
│   ├── switch.md, close.md, review.md ← workflow commands
│   └── style.md                     ← style feedback
│
├── hooks/
│   ├── safety-check.sh              ← blocks force push, rm -rf, push to main
│   └── post-edit.sh                 ← logs every file change
│
└── rules/
    ├── javascript-typescript.md     ← JS/TS style preferences
    └── css.md                       ← CSS/Tailwind preferences

project-templates/                   ← copy to each project
├── README.md.template
├── SPEC.md.template
├── REVIEW.md.template
├── AGENTS.md.template
└── gitignore-additions.txt

---

Per-project setup

cp project-templates/README.md.template  myproject/README.md
cp project-templates/SPEC.md.template    myproject/SPEC.md
cp project-templates/REVIEW.md.template  myproject/REVIEW.md
cp project-templates/AGENTS.md.template  myproject/AGENTS.md
mkdir myproject/.tasks
cat project-templates/gitignore-additions.txt >> myproject/.gitignore

Then open the project in VS Code and start with /plan.

---

Branch naming

The divider creates branches in the format [feature]/[agent]. Example for feature "auth":

• auth/backend
• auth/database
• auth/frontend

Each agent works on its own branch. You merge manually after /review passes. Claude never merges.

---

The two types of context

Personal context (~/.claude/) — who can do what, rules, style preferences. Applies to every project. Never changes per project.

Project context (project root) — README.md, SPEC.md, .tasks/. Specific to this project and feature cycle. Read by every agent at session start.

---

Updating .md files

Nothing updates automatically. If a conversation results in a useful new rule:

1. Ask the agent to read its current .md file
2. It proposes the exact update
3. You paste it in manually

This applies to CLAUDE.md, agent files, style files, README.md, SPEC.md, and task files.