AI Workbench
PT-BR
Back to Home

Setup Wizard

Configure Claude Code, Codex, OpenCode, Zed, and VS Code in a single project. The wizard reads your code, detects the stack, and generates the full structure automatically.

Prerequisites

Required

  • • Node.js 18+ — for the wizard and MCPs via npx
  • • Claude Code — primary orchestrator
  • • Codex CLI — review, delegation, and subagents (npm i -g @openai/codex)
  • • Git — the wizard uses worktrees for isolation

Optional

  • • OpenCode — alternative open-source executor
  • • Zed / VS Code — alternative editors

Step by Step

1

Clone the repository

git clone https://github.com/jdonini/ai-workbench.git
cd ai-workbench
2

Run the wizard

claude ./templates/setup-wizard.md

The wizard asks 12 questions about your project (stack, type, state, team, workflow, and more) and auto-detects what it can.

3

Generated structure

project/
├── CLAUDE.md                          ← Instructions for Claude Code
├── AGENTS.md                          ← Instructions for Codex / OpenCode
├── .mcp.json                          ← Configured MCP servers
│
├── .claude/
│   ├── settings.json                  ← Permissions, effort, hooks
│   ├── .last-setup.json               ← Wizard state
│   ├── hooks/
│   │   ├── session-start.sh
│   │   └── pre-compact.sh
│   ├── rules/
│   │   ├── universal.md
│   │   ├── project-context.md
│   │   ├── decisions.md
│   │   ├── glossary.md
│   │   └── {stack}.md
│   ├── commands/
│   │   ├── start-feature.md
│   │   ├── debug-issue.md
│   │   └── review-code.md
│   ├── skills/
│   │   ├── feature-start-detector/
│   │   ├── multi-stack-detector/
│   │   ├── dependency-guard/
│   │   └── auto-dream/
│   └── agents/
│       ├── code-reviewer.md
│       ├── language-guide-{stack}.md
│       └── error-detective.md
│
├── .codex/
│   ├── config.toml                    ← Model, sandbox, approval
│   ├── hooks.json                     ← Lifecycle hooks
│   ├── commands/
│   │   ├── start-feature.md
│   │   ├── debug-issue.md
│   │   └── review-code.md
│   ├── rules/ → link para .claude/rules/
│   └── agents/
│       ├── explorer.toml
│       ├── reviewer.toml
│       └── {stack}-worker.toml
│
├── .agents/
│   └── skills/                        ← Skills (open agent skills standard)
│       ├── feature-start-detector/
│       ├── multi-stack-detector/
│       ├── dependency-guard/
│       ├── code-reviewer/
│       └── error-detective/
│
├── .opencode/
│   ├── rules/ → link para .claude/rules/
│   └── commands/
│       ├── start-feature.md
│       ├── debug-issue.md
│       └── review-code.md
│
└── .claude.local/                     ← Personal overrides (gitignored)

Required Plugin

In Phase 0, the wizard installs superpowers. It implements the Spec-Driven Development workflow:

superpowers

Workflow SDD
1

Brainstorming

Align ideas and scope before coding. Explore intent, requirements, and design.

2

Planning with Files

Granular planning in files (`task_plan.md`, `findings.md`, `progress.md`). Independent tasks with verification criteria.

3

Test-Driven Development

Red → Green → Refactor. Write the test first, then the minimum code to pass.

4

Systematic Debugging

Hypothesis → investigation → surgical fix. Do not refactor while debugging.

5

Verification Before Completion

Evidência antes de afirmar. Rode testes e lint antes de declarar sucesso.

6

Receiving Code Review

Technical rigor when receiving feedback. Verify before implementing suggestions.

7

Finishing a Development Branch

Merge, PR, or cleanup. Clean commits and a branch ready for review.

MCP Servers

The wizard configures MCP servers via .mcp.json. Each server connects automatically when the project opens.

mem0

Persistent semantic memory. Entities, decisions, and preferences are stored and retrieved automatically.

Requires MEM0_API_KEY

napkin

Cumulative project runbook. Learns from each interaction and organizes knowledge by prioritized categories in .claude/napkin.md.

context7

Real-time library and framework documentation. Fetches up-to-date context for any API.

sequential-thinking

Step-by-step reasoning for complex problems. Helps the model think in a structured way.

postman

Integration with Postman collections to test and document APIs directly from the development context.

postgresql

Database access for queries, schema exploration, and debugging directly from context.

Configuration .mcp.json

{
  "mcpServers": {
    "mem0": {
      "command": "uvx",
      "args": ["mem0-mcp-server"],
      "env": {
        "MEM0_API_KEY": "$MEM0_API_KEY",
        "MEM0_DEFAULT_USER_ID": "$MEM0_DEFAULT_USER_ID"
      }
    },
    "context7": {
      "command": "npx",
      "args": ["-y", "@upstash/context7-mcp@latest"],
      "env": { "CONTEXT7_API_KEY": "$CONTEXT7_API_KEY" }
    },
    "sequential-thinking": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-sequential-thinking"]
    },
    "postman": {
      "command": "npx",
      "args": ["-y", "@postman/postman-mcp-server"],
      "env": { "POSTMAN_API_KEY": "$POSTMAN_API_KEY" }
    },
    "postgresql": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-postgres", "postgresql://..."]
    }
  }
}

Next Step

Now that the environment is configured, explore the standard development workflow.

View Workflow