- Implement multi-agent orchestration system for parallel AI tasks - Add role-based AI assignment (coder, reviewer, critic, architect, etc) - Create predefined workflows (code-review, implement-feature, optimize, debug) - Add local LLM support with auto-detection (Ollama, LocalAI, etc) - Implement async job management with idle timeout - Add authentication system with OAuth2 PKCE flow - Create comprehensive test suite with integration tests - Add headless Chrome testing for auth flows - Create mock AI server for testing all endpoints - Add visual test runners and demo scripts - Update CI/CD pipeline for all services - Rename CLI from 'hanzo-dev' to 'dev' for simplicity - Update repository references to github.com/hanzoai/dev Key features: - Parallel execution of multiple AI agents - Git worktree support for isolated development - Universal context sync across tools - Secure API key management - Comprehensive workflow system
3.6 KiB
3.6 KiB
Dev CLI Examples
Basic Usage
# Quick start
dev login
dev init
# Run any AI tool
dev run claude "write a Python web server"
dev run aider "fix the failing tests"
dev run gemini "explain this codebase"
Multi-Agent Workflows
Code Review
# Review current changes
dev review
# Review specific files
dev review src/api.js src/auth.js
# Deep security-focused review
dev workflow custom-review "$(git diff)"
Feature Implementation
# Full feature implementation workflow
dev workflow implement-feature "add OAuth2 authentication"
# This runs:
# 1. Claude designs the architecture
# 2. Aider implements code + Codex writes tests + Gemini writes docs (parallel)
# 3. Claude & Gemini review the implementation
Custom Multi-Agent Tasks
# Specify which tool handles what
dev multi "optimize database queries" \
--coder claude \
--reviewer gemini \
--critic codex
# Mix local and cloud models
dev multi "refactor authentication system" \
--coder local:codellama \
--reviewer claude \
--critic gemini
Local LLM Integration
# Use Ollama (auto-detected)
dev run local-llm "explain this function" --model llama3
# Use specific local provider
dev multi "implement caching layer" \
--local llama3:latest \
--reviewer claude
# Configure custom local LLM
echo '{
"providers": [{
"name": "my-llm",
"endpoint": "http://192.168.1.100:8080",
"models": ["my-model"],
"defaultModel": "my-model",
"apiFormat": "openai"
}]
}' > ~/.dev/local-llm.json
dev run local-llm "generate tests" --provider my-llm
Async and Parallel Execution
# Long-running task in background
dev run claude "refactor entire codebase to TypeScript" --async
# Returns: Job ID: abc123...
# Check progress
dev status abc123
# Send additional instructions
dev input abc123 "focus on the API layer first"
# Run multiple agents in parallel branches
dev run claude "implement user service" --worktree &
dev run aider "implement auth service" --worktree &
dev run openhands "implement notification service" --worktree &
# Check all worktrees
dev worktree list
Advanced Workflows
Performance Optimization
dev workflow optimize "SELECT * FROM users WHERE created_at > '2024-01-01'"
# Claude and Codex analyze in parallel, then Aider implements the best solution
Debugging
dev workflow debug "app crashes when processing large files"
# Multiple agents diagnose the issue from different angles
Custom Workflow
# Create your own workflow
cat > ~/.dev/workflows/my-workflow.json << 'EOF'
{
"name": "my-workflow",
"description": "My custom workflow",
"steps": [
{
"name": "analyze",
"agents": [
{ "role": "architect", "tool": "claude" },
{ "role": "critic", "tool": "local-llm", "model": "llama3" }
],
"parallel": true
},
{
"name": "implement",
"agents": [
{ "role": "coder", "tool": "aider" }
],
"parallel": false
}
]
}
EOF
dev workflow my-workflow "implement new feature"
Tips and Tricks
# List all workflows
dev workflow list
# Compare all tools at once
dev compare "how would you implement a rate limiter?"
# Quick code review before commit
git add .
dev review # Reviews staged changes
git commit -m "feat: add rate limiting"
# Use with git hooks
echo 'dev review --type quick' >> .git/hooks/pre-commit
# Pipe output to other tools
dev run claude "generate API spec" | swagger-cli validate
# Use in scripts
#!/bin/bash
for file in src/*.js; do
dev review "$file" >> review-report.md
done