Robust Claude Project Kickoff

Starting a project with Claude isn't just "write code." A thoughtful kickoff includes discovery, tooling configuration, and context setup. Here's a systematic approach.

1. Discovery Prompts

Before writing code, have Claude ask clarifying questions. Use /init or start with a meta-prompt:

Initial prompt

I want to build [your idea]. Before we start coding:

1. What clarifying questions do you have about requirements?
2. What technical decisions need my input (language, framework, hosting)?
3. What constraints should I be aware of (budget, timeline, existing systems)?
4. What's the MVP scope vs. future enhancements?

Ask me these questions before proposing any architecture.

bash

2. Generate CLAUDE.md

After discovery, have Claude generate project context that persists across sessions:

CLAUDE.md template prompt

Based on our discussion, generate a CLAUDE.md file that includes:

- Project overview and goals
- Tech stack decisions and why
- Coding conventions (naming, patterns, structure)
- Testing requirements
- Commands to run (build, test, deploy)
- What NOT to do (anti-patterns to avoid)

bash

3. Configure Hooks

Set up .claude/settings.json with project-specific automation:

.claude/settings.json

{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "Edit|Write",
        "hooks": [
          { "type": "command", "command": "npm run format -- $(jq -r '.tool_input.file_path')" }
        ]
      }
    ],
    "Stop": [
      {
        "matcher": "",
        "hooks": [
          { "type": "command", "command": "npm run typecheck" }
        ]
      }
    ]
  },
  "permissions": {
    "allow": [
      "Bash(npm run *)",
      "Bash(git status)",
      "Bash(git diff *)"
    ]
  }
}

bash

4. MCP Servers

Configure Model Context Protocol servers for external integrations:

~/.claude/settings.json (user-level)

{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": { "GITHUB_TOKEN": "your-token" }
    },
    "postgres": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-postgres", "postgresql://..."]
    },
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/docs"]
    }
  }
}

bash

5. Leverage Subagents

Explore agent - Codebase discovery and understanding
Plan agent - Architecture design before implementation
Bash agent - Complex shell operations and git workflows
claude-code-guide agent - Claude Code feature questions

Trigger Plan mode for non-trivial features to get architecture approval before coding.

Full kickoff workflow

Run discovery → generate CLAUDE.md → configure hooks → set up MCP servers → scaffold project structure → commit initial setup. Each session after starts with full context.

1. Discovery Prompts

2. Generate CLAUDE.md

3. Configure Hooks

4. MCP Servers

5. Leverage Subagents

Related Snippets