Claude Code Project Template v3.4

A production-ready template for Claude Code that turns it into a team of specialized agents with persistent memory, domain-specific skills, and automated workflows. Built for Turborepo + React/Vite + Supabase monorepos.

Open Specs compliant — all 12 skills follow the Agent Skills open standard, making them portable to GitHub Copilot, OpenAI Codex CLI, Cursor, and 20+ other compatible tools.

Why this template?

Claude Code out of the box is a general-purpose assistant. This template transforms it into an opinionated engineering team that:

• Plans before coding — evaluates scope, risks, and affected workspaces before writing a single line
• Follows consistent patterns — every feature, form, query, and test follows documented conventions
• Learns from mistakes — a self-improvement loop captures lessons and prevents repeated errors
• Verifies its own work — runs type-check, lint, and tests before marking anything as done
• Manages context efficiently — uses subagents and progressive disclosure to stay within token limits

What’s inside

.claude/
├── agents/                                  # 4 specialized agents with persistent memory
│   ├── implementer.md                       # Builds features following project conventions
│   ├── code-reviewer.md                     # Reviews code for quality, UI, forms, a11y
│   ├── researcher.md                        # Investigates packages/APIs with anti-hallucination
│   └── test-engineer.md                     # Writes and runs tests (Vitest + Testing Library)
│
├── skills/                                  # 12 skills (Open Specs compliant)
│   ├── scaffold-feature/                    # Generate feature file structure
│   │   ├── SKILL.md
│   │   └── references/CODE-PATTERNS.md
│   ├── supabase-patterns/                   # Validated Supabase patterns
│   │   ├── SKILL.md
│   │   └── references/ADVANCED.md, TANSTACK.md
│   ├── ui-patterns/                         # UI: layouts, tables, feedback, pages
│   │   ├── SKILL.md
│   │   └── references/LAYOUTS.md, DATA-TABLES.md, FEEDBACK.md, PAGES.md
│   ├── forms-validation/                    # Zod + React Hook Form + shadcn/ui
│   │   ├── SKILL.md
│   │   └── references/PATTERNS.md, WIZARD.md
│   ├── code-quality/                        # ESLint, Prettier, tsconfig, Vite configs
│   │   ├── SKILL.md
│   │   └── references/CONFIGS.md
│   ├── commit/SKILL.md                      # Auto-bump patch + conventional commits
│   ├── deploy-check/SKILL.md                # Pre-merge verification checklist
│   ├── new-feature/SKILL.md                 # Full feature pipeline
│   ├── evaluate/SKILL.md                    # Scope, effort, and risk assessment
│   ├── review/SKILL.md                      # Code quality review
│   ├── research/SKILL.md                    # Technical research with source verification
│   └── status/SKILL.md                      # Project status report
│
├── docs/                                    # Agent-generated docs (created during use)
└── settings.json                            # 49 allow rules, deny list, hooks

tasks/
├── todo.md                                  # Active task plan with verifiable checklist
└── lessons.md                               # Self-improvement loop: patterns from mistakes

CLAUDE.md                                    # Project brain: conventions, principles, directives
.env.example                                 # Environment variables template
.github/workflows/ci.yml                     # PR → type-check + lint + test + build
.github/workflows/deploy.yml                 # main → Supabase migrations + Vercel deploy

Quick start

# 1. Create your monorepo
npx create-turbo@latest my-project --pm pnpm
cd my-project

# 2. Copy the template into your project
cp -r /path/to/template/.claude .
cp -r /path/to/template/.github .
cp -r /path/to/template/tasks .
cp /path/to/template/CLAUDE.md .
cp /path/to/template/.env.example .

# 3. Init Supabase
npx supabase init

# 4. Customize CLAUDE.md for your project
# Edit the "DIRECTIVAS ESPECÍFICAS" section

# 5. Start Claude Code
claude

Daily workflow

/status                                  # Project overview + version
/evaluate user invitation system         # Assess scope before building
/new-feature invitations                 # Full pipeline: scaffold → implement → test → review → commit
/review                                  # Quality check
/commit feat: add invitations            # Auto-bump patch version + conventional commit
/deploy-check                            # Pre-merge verification

Tech stack

This template is configured for:

Agents

Four specialized agents, each with persistent memory across sessions:

Every agent starts by reading tasks/lessons.md and ends by updating it if corrections were made.

Skills

All 12 skills follow the Agent Skills open standard:

Progressive disclosure

Skills use a 3-level loading strategy to minimize token usage:

1. Metadata (~100 tokens) — name + description loaded at startup for all 12 skills
2. Instructions (~200 tokens) — SKILL.md body loaded only when the skill is activated
3. References (on-demand) — detailed files loaded only when the specific task requires them

Five skills include references/ directories with 10 total reference files that Claude loads selectively.

Behavioral principles

CLAUDE.md encodes six workflow orchestration directives:

1. Planning by default — enter planning mode for any non-trivial task; write specs upfront
2. Subagent strategy — delegate research, exploration, and parallel work to subagents
3. Self-improvement loop — after every user correction, capture the pattern in tasks/lessons.md
4. Verification before completion — never mark done without proving it works; ask “would a Staff Engineer approve this?”
5. Demand elegance — pause on non-trivial changes and ask “is there a more elegant way?”; no hacky patches
6. Autonomous error fixing — when receiving a bug report, fix it without asking for hand-holding

And three core principles that govern all work:

• Simplicity first — affect the minimum code necessary
• No laziness — find root causes, no temporary fixes, senior developer standards
• Minimal impact — changes only touch what’s needed, avoid introducing errors

Open Specs compliance

Every skill meets the agentskills.io specification:

Claude Code extensions (model, context: fork, argument-hint) are preserved for Claude Code users and silently ignored by other compatible tools.

Anti-hallucination

Multiple layers prevent Claude from inventing APIs, packages, or patterns:

1. CLAUDE.md — strict verification protocol for all external references
2. Researcher agent — verifies against official docs before recommending
3. Skills with warnings — each skill states what to verify and what NOT to assume
4. Confidence levels — agents report HIGH/MEDIUM/LOW on recommendations
5. Honesty principle — “I don’t know” is a valid configured response
6. Hooks + deny list — settings.json blocks dangerous commands

Customization

The template is designed to be forked and customized:

• CLAUDE.md → edit “DIRECTIVAS ESPECÍFICAS” for your project’s domain rules
• Skills → add new skills in .claude/skills/[name]/SKILL.md following the Open Specs format
• Agents → modify agent prompts in .claude/agents/ to match your team’s conventions
• settings.json → adjust allow/deny rules for your toolchain
• tasks/ → the self-improvement loop works automatically as you use the template

Adapting the stack

While this template targets React + Supabase, the architecture (agents, skills, principles, task management) is stack-agnostic. To adapt:

1. Replace supabase-patterns with your backend’s patterns
2. Update scaffold-feature file structure for your framework
3. Adjust code-quality configs for your toolchain
4. Keep everything else — the workflow orchestration, agents, and principles work with any stack

Version history

License

MIT

Author

[email protected] — Built with Claude Code for Claude Code.