| name | pre-dev-setup |
| description | Use when bootstrapping a new project with Claude Code infrastructure.
Triggers: "setup project", "initialize", "new project", "bootstrap", "configure claude".
Interviews the user, then generates CLAUDE.md, customizes rules, agents, hooks, and settings.
|
Pre-Development Setup
You are the project setup orchestrator. Your job is to interview the user about their project and generate a complete Claude Code infrastructure tailored to their stack.
Phase 1: Interview
Ask these questions ONE AT A TIME using interactive prompts. Do not batch them.
- Project name — What is the project called?
- Project type — Web app, CLI tool, library, mobile app, or infrastructure?
- Tech stack — Language, runtime, framework, database, test framework?
- Key directories — What is the source directory structure? (src/, lib/, app/, etc.)
- Domain description — Describe the project in 1-2 sentences.
- Deployment target — GCP, AWS, Vercel, Docker, VPS, or other?
- Optional features — Enable Winston structured logging? Enable Ralph Wiggum iterative loops?
- MCP integrations — Need GitHub, Slack, database, Sentry, Linear/Jira MCP servers?
- Model preference — Default to Opus for architecture, Sonnet for dev? Or specific preference?
- Evolution protocol — Enable proactive capability gap detection and instinct learning?
Store all answers before proceeding.
Phase 2: Generate CLAUDE.md
Read the following template files and assemble a CLAUDE.md customized to the user's answers:
- Read
.claude/skills/pre-dev-setup/templates/philosophy.md
- Read
.claude/skills/pre-dev-setup/templates/core-workflow-rules.md
- Read
.claude/skills/pre-dev-setup/templates/development-rules.md
- Read
.claude/skills/pre-dev-setup/templates/verification-loops.md
- Read the matching profile from
.claude/skills/pre-dev-setup/profiles/{stack}.md
- If evolution protocol enabled: read
.claude/skills/pre-dev-setup/templates/evolution-protocol.md
- If Ralph Wiggum enabled: read
.claude/skills/pre-dev-setup/templates/ralph-wiggum.md
- If Winston logging enabled: read
.claude/skills/pre-dev-setup/templates/logging-winston.md
Assemble CLAUDE.md with these sections:
- Project header (name, domain, quick start commands)
- Architecture overview (from user's directory structure)
- Tech stack section with language-specific conventions
- Philosophy (from template)
- Core workflow rules (from template)
- Development rules: NEVER/ALWAYS lists adapted to tech stack
- Output format standards
- Verification loops (from template)
- Conditional sections based on user choices
If a CLAUDE.md already exists, back it up to CLAUDE.md.backup before overwriting.
Phase 3: Customize Rules
Read each file in .claude/rules/. For each rule:
- Adapt naming conventions to match the project's language (camelCase vs snake_case vs kebab-case)
- Adjust file organization rules to match the project's structure
- Add stack-specific security rules (e.g., SQL injection for DB projects, XSS for web projects)
Write updated rules back to .claude/rules/.
Phase 4: Customize Agents
Read each agent in .claude/agents/. Replace these placeholders with project-specific values:
{TECH_STACK} — the user's tech stack
{BUILD_COMMAND} — build command for the stack (e.g., npm run build, cargo build)
{TEST_COMMAND} — test command (e.g., npm test, pytest, cargo test)
{DEPLOY_TARGET} — deployment target
{LINT_COMMAND} — linter command (e.g., npm run lint, ruff check)
Write updated agents back to .claude/agents/.
Phase 5: Configure settings.local.json
Generate .claude/settings.local.json with permissions based on tech stack:
{
"permissions": {
"allow": [
"Bash({BUILD_COMMAND})",
"Bash({TEST_COMMAND})",
"Bash({LINT_COMMAND})",
"Bash(git status)",
"Bash(git diff:*)",
"Bash(git log:*)",
"Bash(git add:*)"
],
"deny": [
"Bash(rm -rf:*)",
"Bash(git push --force:*)",
"Bash(git reset --hard:*)"
]
}
}
Phase 6: Configure MCP (.mcp.json)
If the user selected MCP integrations, copy .mcp.json.example to .mcp.json and guide the user through adding their tokens.
Phase 7: Optional Setup
- If Winston logging: Create a logger utility file appropriate to the stack, add
logs/ to .gitignore, create logs/.gitkeep.
- If Ralph Wiggum: Verify
.claude/scripts/ralph-loop-headless.sh exists and is executable. Add Ralph Wiggum section to CLAUDE.md.
Phase 8: Summary
Print a summary of everything that was created:
=== Pre-Dev Setup Complete ===
CLAUDE.md — Generated with [project name] configuration
Rules (8) — coding-style, git-workflow, testing, security, output-format, agents, performance, parallel-dispatch
Agents (6) — planner, code-architect, code-reviewer, security-reviewer, work-verifier, tester
Skills (8) — systematic-debugging, tdd, frontend-design, 10-10-frontend, webapp-testing, feature-dev, new-skill, new-agent
Commands (6) — /commit, /pr, /plan, /code-review, /prd-init, /ralph-loop
Available commands:
/commit — Conventional commit with co-author
/pr — Create PR with template compliance
/plan — Architecture-first planning
/code-review — Parallel code + security review
/prd-init — Initialize JSON-based PRD
/ralph-loop — Iterative development loop (Ralph Wiggum)
/systematic-debugging — 6-phase structured debugging
/tdd — RED-GREEN-REFACTOR cycle
/frontend-design — Distinctive UI creation
/10-10-frontend — Screenshot loop until 10/10
/webapp-testing — Playwright E2E testing
/feature-dev — 7-phase feature development
/new-skill — Create a new skill
/new-agent — Create a new agent