Back to Projects

Dev Template

Developer ToolsOngoingSolo project2026
TypeScriptClaude CodeCrewAIVitestESLint 9HuskyGitleaksGitHub ActionsChangesetspnpm
Dev Template
6 specialized AI sub-agents for codebase research
30+ slash commands for plan-driven development
Three-layer git hook protection (secrets, lint, local CI)
Persistent AI knowledge base across sessions
Zero application code — pure development infrastructure

Dev Template - Agentic Development Starter

Open Source Template for Building with AI Agents

Project Overview

Dev Template is a tooling-only starter template designed for building startups with a Next.js + Supabase + Vercel stack. It's explicitly not a scaffold — it contains zero application code. Instead, it provides a fully pre-configured development infrastructure layer that you fork and build on top of. The key differentiator is its deep optimization for AI-assisted (agentic) development via Claude Code, with 6 specialized sub-agents and 30+ slash commands.

The Problem: Setting up proper development infrastructure — CI/CD, git hooks, testing, linting, secret scanning, AI agent configuration — takes days of work that every team repeats from scratch. And when you're building with AI coding assistants, there's no standard way to configure them for consistent, high-quality output.

The Solution: Fork this template and immediately get a production-grade development environment with a complete agentic development framework built in.

The Agentic Development System

This is the most distinctive part of the template. The .claude/ directory contains a complete framework for AI-assisted development.

6 Specialized Sub-Agents

Each agent runs on the Sonnet model and has a focused responsibility:

Agent Purpose
Codebase Locator "Super Grep" — finds WHERE code lives without reading contents
Codebase Analyzer Analyzes HOW code works with file:line precision
Pattern Finder Finds reusable patterns with concrete code examples
Web Researcher Web research for up-to-date external information
Thoughts Analyzer Extracts insights from research documents
Thoughts Locator Discovers relevant documents in the knowledge base

All agents are explicitly instructed to be "documentarians, not critics" — they describe what exists without suggesting improvements, preventing AI from going off-track during research.

30+ Slash Commands

Key workflow commands include:

  • /create_plan — Interactive plan creation that spawns parallel sub-agents for codebase research, generates phased implementation plans with success criteria
  • /implement — End-to-end GitHub issue implementation: reads the issue, researches codebase, creates branch, generates plan, implements phase-by-phase, commits, and creates PR
  • /implement_plan — Executes a plan phase-by-phase with verification, pausing for manual testing between phases
  • /research_codebase — Comprehensive codebase documentation using parallel sub-agents
  • /feature — Turns plain-language feature requests into codebase-aware GitHub issues
  • /debug — Investigation-only analysis of logs, database state, and git history
  • /create_handoff / /resume_handoff — Transfers work context between sessions via structured documents

Persistent Knowledge Base

The thoughts/ directory serves as long-term memory for AI agents across sessions, organized into plans, research, tickets, handoffs, and PR descriptions.

Development Infrastructure

Three-Layer Git Hook Protection

Every commit passes through:

  1. Pre-commit: Gitleaks secret scanning + lint-staged (ESLint + Prettier)
  2. Commit-msg: Commitlint validates conventional commit format
  3. Pre-push: Runs the full GitHub Actions CI pipeline locally via act in Docker

CI/CD Pipeline

  • ESLint + Prettier validation
  • TypeScript strict type checking
  • Vitest unit tests with 60% coverage thresholds
  • Gitleaks secret detection (with dual-environment support for GitHub Actions and local act)
  • Changeset verification on PRs

Custom Secret Scanning

Extended Gitleaks configuration with custom rules for Supabase service keys, Vercel API tokens, and NextAuth secrets — with smart allowlists for test files and examples.

Technology Stack

  • TypeScript 5.7+ (strict mode)
  • ESLint 9 (flat config) + Prettier 3.8
  • Vitest 3.0 + React Testing Library (60% coverage thresholds)
  • Husky 9.1 (3 git hooks)
  • commitlint (conventional commits)
  • Changesets (semantic versioning)
  • Gitleaks (secret scanning)
  • nektos/act (local CI via Docker)
  • pnpm (enforced)

How to Use

  1. Fork the repository
  2. Layer your Next.js application on top
  3. The entire development infrastructure — CI/CD, hooks, testing, AI agents — is ready to go

Links:

Open source under MIT license. Contributions welcome.

View Code Repository
Back to All Projects