Skip to content

jleechanorg/ai_dev_recs

BranchesTags

Repository files navigation

AI Development Recommendations

Production-tested AI development toolchain for terminal-first workflows

📑 Table of Contents

📋 Executive Summary

Quick Reference: All Recommended Tools

1. AI Coding Tools (CLIs & IDEs)

Tool What It Is Why We Recommend It
Claude Code CLI Anthropic's official command-line AI assistant with 200K context window Terminal-first workflow enables scripting and automation. Extensible via MCP servers. Supports 80+ custom slash commands and multi-agent coordination. The foundation of the entire toolchain.
Cursor VS Code fork with native AI pair programming Best-in-class autocomplete (multi-line, contextual). Chat with codebase using @ symbols. Cmd+K for inline edits. $20/mo.
Codex CLI Alternative CLI AI assistant focused on precision Excellent for precision work and verification. Provides detailed, accurate responses when you need high-quality output without iteration.
Antigravity Web-based AI IDE with generous token limits Unconstrained usage of Opus 4.6 and other premium models. Great for heavy workloads. Warning: Agent manager is flaky and crashes frequently - save work often.

2. Utilities & Tools

Tool What It Is Why We Recommend It
AI Usage Tracker Token usage and cost tracking for Claude Code and Codex CLI Essential for monitoring AI spend across platforms. Combines usage data from multiple sources into unified reports with daily averages and cache efficiency metrics. Commands: ai-usage-tracker (pip) or ai-usage-tracker-js (npm).
Warp Terminal Modern terminal with AI features and enhanced UX Built-in LLM inference for command suggestions and explanations. Renamable tabs for organization (critical when running multiple AI agents). Block-based output and command palette. Makes multi-agent workflows manageable. Free tier available.

3. MCP Servers

MCP Server What It Is Why We Recommend It
Gemini MCP Google's LLM accessed via MCP integration Provides structured output with JSON mode, code execution capabilities, and reliable fallback chains for production workloads. Multi-model coordination.
beads MCP-integrated task tracking system AI agents can read/write tasks in real-time. Superior to Jira/Linear for AI-first workflows.
grok X.ai's real-time intelligence LLM Real-time data and current events. Complements Claude and Gemini for fresh information.
perplexity-ask Research and web search via Perplexity Fast research with citations. Better than manual Google searches for technical questions.
sequential-thinking Enhanced reasoning chains Improves complex problem-solving by breaking down multi-step reasoning.
chrome-superpower Browser automation via Chrome DevTools Protocol Automate web interactions, testing, and scraping directly from AI workflows.
context7 Library documentation lookup Instant access to up-to-date docs for any library without leaving terminal.
mcp_mail Agent-to-agent messaging system Enables AI agents to communicate, coordinate, and share context. Essential for multi-agent workflows and task handoffs.

4. Slash Commands

Command What It Is Why We Recommend It
Obra Superpowers Browser automation and control via Chrome extension Direct browser control from Claude Code. Navigate, click, fill forms, screenshot - all from your AI workflow. Powerful for web testing and automation.
/copilot Autonomous PR processing command Handles 100+ comment PRs automatically. 4-phase workflow with priority-based triage (CRITICAL → BLOCKING → IMPORTANT → ROUTINE). Requires gh CLI.
/fixpr Intelligent PR fix automation Analyzes failures, identifies root cause, makes minimal code changes. No gold-plating. Integrated verification. Requires gh CLI.
/redgreen Test-driven development workflow Enforces red-green pattern with evidence-based testing. Structured artifacts for reproducibility.
/fake Automatic code quality gate (hook) Runs after EVERY Write operation. Catches 15% of code with issues (fake selectors, placeholder code).
/execute General task execution command Versatile command for running complex multi-step tasks. Heavy daily usage across all workflows.

Core Tools (Daily Use)

1. Claude Code CLI

What it is: Anthropic's official command-line AI assistant with 200K context window

Why valuable:

  • Terminal-first workflow (scriptable, automatable)
  • Extensible via MCP servers (10 active integrations)
  • Custom commands (80+ slash commands for specific workflows)
  • Multi-agent coordination support

How to install:

# Install Claude Code
curl -fsSL https://cli.claude.ai/install.sh | sh

# Set API key
export ANTHROPIC_API_KEY="sk-ant-..."

Detailed docs: TOOLCHAIN.md, SETUP_GUIDE.md

2. Gemini (Google)

What it is: Primary LLM backend via MCP integration

Why valuable:

  • 1,440 MCP hits + 1,114 commit mentions (highest usage)
  • Structured output with JSON mode
  • Code execution capabilities
  • Fallback chains for reliability

How to install:

# Via MCP server (gemini-cli-mcp)
# See MCP Servers section below

Detailed docs: TOOLCHAIN.md, WORKFLOWS.md

3. beads (bd)

What it is: MCP-integrated task tracking system

Why valuable:

  • AI can read/write tasks in real-time
  • Used MORE than traditional issue trackers (259 commits)
  • Binary accessed daily (last access: yesterday)

How to install:

# Already installed in the reference setup
bd list
bd create "Task description"

Detailed docs: TOOLCHAIN.md, WORKFLOWS.md

AI Assistants

4. Cursor

What it is: VS Code fork with native AI pair programming

Why valuable:

  • Best-in-class autocomplete (multi-line, contextual)
  • Chat with codebase using @ symbols
  • Cmd+K for inline edits
  • Active agent branches (cursor/*)

How to install:

# Download from https://cursor.sh/
# Install and sign in

Cost: $20/mo (Pro)

Detailed docs: SETUP_GUIDE.md

5. Warp Terminal

What it is: Modern terminal with AI features and better UX

Why valuable:

  • Built-in LLM inference - Get command suggestions and explanations without leaving the terminal
  • Renamable tabs - Critical for organizing multiple AI agent sessions (e.g., "copilot-agent", "cursor-build", "claude-review")
  • Block-based output - Each command output is a distinct block, easier to scan when running long workflows
  • Command palette - Search history and saved commands
  • Primary terminal for daily work - Makes multi-agent orchestration manageable

How to install:

# Download from https://www.warp.dev/

Cost: Free

Detailed docs: SETUP_GUIDE.md

6. Antigravity & Codex

What they are: Alternative AI IDEs for specialized tasks

Why valuable:

  • Antigravity: Experimental features (web-based IDE)
  • Codex: Precision and verification (CLI tool)
  • Active agent branches (codex/*)

How to install:

# Download from official sites

Detailed docs: TOOLCHAIN.md

MCP Servers

7. MCP Ecosystem (10 Active Servers)

What it is: Model Context Protocol servers for multi-AI coordination

Why valuable:

  • 1,285 mcp-cli uses in last week alone!
  • Multi-model coordination (Gemini + Grok + Perplexity + Claude)
  • Browser automation (chrome-superpower, claude-in-chrome)
  • Research acceleration (context7 for docs, perplexity for search)

How to install:

npm install -g @modelcontextprotocol/cli
mcp-cli servers  # Verify active servers

Active servers:

  1. gemini-cli-mcp (1,440 hits) - Gemini API
  2. grok (670 hits) - X.ai real-time intelligence
  3. perplexity-ask (661 hits) - Research/search
  4. sequential-thinking (649 hits) - Enhanced reasoning
  5. beads (539 hits) - Task tracking
  6. chrome-superpower (407 hits) - Browser automation
  7. context7 (341 hits) - Library docs
  8. claude-in-chrome - Chrome extension
  9. mcp_mail (28 hits) - Agent messaging
  10. plugin_superpowers-chrome - Obra Superpowers

Detailed docs: WORKFLOWS.md, SETUP_GUIDE.md

Automation Commands

8. /copilot

What it is: Autonomous PR processing command

Why valuable:

  • Handles 100+ comment PRs automatically
  • 4-phase workflow (Critical scan → Categorize → Implement → Verify)
  • Priority-based triage (CRITICAL → BLOCKING → IMPORTANT → ROUTINE)

How to use:

/copilot  # Processes all open PR comments

Detailed docs: WORKFLOWS.md

9. /fixpr

What it is: Intelligent PR fix automation

Why valuable:

  • Analyzes failures, identifies root cause
  • Minimal code changes (no gold-plating)
  • Integrated verification

How to use:

/fixpr        # Fix current PR
/fixpr 123    # Fix specific PR number

Detailed docs: WORKFLOWS.md

10. /redgreen (TDD)

What it is: Test-driven development workflow

Why valuable:

  • Red-green pattern (failing test → minimal code → pass → refactor)
  • Evidence-based testing with structured artifacts
  • 17.4% of commits are tests

How to use:

/redgreen  # or /tdd

Detailed docs: WORKFLOWS.md

11. /fake (Automatic Hook)

What it is: Automatic code quality gate that runs after EVERY Write operation

Why valuable:

  • Proven track record: Catches 15% of code with issues (speculative logic, fake selectors, placeholder code)
  • Real catches: Found 9+ violations in automate_codex_update branch (fake UI selectors, placeholder URLs)
  • Zero effort: 343 automatic checks in 30 days, no manual invocation
  • Low cost: ~$5-10/month in API calls, prevents hours of debugging

How to setup:

# Environment variable
export SMART_FAKE_TIMEOUT=180  # 3 minutes

# Hook runs automatically via PostToolUse
# See WORKFLOWS.md for full setup

Detailed docs:

12. Other Commands

Active commands with evidence:

  • /execute (6,332 refs) - General task execution
  • /cerebras - Fast code generation
  • /commentreply (4,007 refs) - Comment automation
  • /research (2,213 refs) - Research patterns
  • /think (1,743 refs) - Reasoning
  • /pr (1,821 refs) - PR creation
  • /orch (1,397 refs) - Multi-agent orchestration

Detailed docs: TOOLCHAIN.md

📊 Quick Stats

Metric Value
Git commits analyzed 19,044 (last 6 months)
Custom slash commands 80+ total, 18 heavily used
MCP servers active 10 of 17 installed
AI assistants 6 (Claude Code, MiniMax 2.5, Cursor, Antigravity, Codex, Warp)
Testing commits 3,308 (17.4% of all work)
PR automation 7,950 /copilot invocations
Multi-agent branches 4 agent prefixes (copilot/, cursor/, codex/, claude/)
mcp-cli usage 1,285 uses in last week
Docker mentions 475 in last week
/fake quality gate 343 automatic checks (30 days), 15% catch rate
Monthly AI cost ~$40-60

🚀 The Complete Stack

See TOOLCHAIN.md for complete tool inventory with evidence scores.

Tier 1 (Daily): Claude Code, pytest, gh CLI, Gemini, beads, tmux

Tier 2 (Heavy): /copilot, /fixpr, /cerebras, Cursor, MCP servers, comment automation

Tier 3 (Regular): Docker, Playwright, cloud services, /secondo, test generation

💡 Unique Innovations

1. 12-Step Multi-AI Workflow

Complete development lifecycle leveraging web + local AI tools.

Source: LinkedIn Post

Key insight: "Focus on AI review/quality as much as generation"

Details: WORKFLOWS.md

2. Multi-Agent Orchestration

4 AI agents (copilot, cursor, codex, claude) working autonomously in parallel.

Details: WORKFLOWS.md

3. Automatic Quality Gates

/fake runs after EVERY file write (343 times last week).

Details: WORKFLOWS.md

4. MCP-First Architecture

10 servers, 1,285 uses in last week, multi-model coordination.

Details: WORKFLOWS.md

5. Evidence-Based Testing

Full paths, structured bundles, 17.4% of commits are tests.

Details: WORKFLOWS.md

📚 Documentation

🚀 Getting Started

Minimal Setup (30 minutes)

# 1. Claude Code CLI
npm install -g @anthropic-ai/claude-code
export ANTHROPIC_API_KEY="sk-ant-..."

# 2. GitHub CLI
brew install gh
gh auth login

# 3. Basic tools
brew install tmux jq

# 4. Test
claude --version

Full Setup (4-8 hours)

See SETUP_GUIDE.md for complete installation of all tools.

Learning Path (4 weeks)

Week 1: Claude Code + basic commands Week 2: Add 2-3 MCP servers (sequential-thinking, gemini, perplexity) Week 3: Testing + quality automation (pytest, /fake hook) Week 4: Multi-agent orchestration

💰 Cost Breakdown

Current Stack (~$40-60/mo)

  • Claude Code Paid: $20/mo (200K context)
  • Cursor Pro: $20/mo
  • Optional: Perplexity Pro $20/mo
  • Alternative: MiniMax 2.5 $20/mo - higher capacity than Claude (see CLAUDE_CODE_MINIMAX_2.5_SETUP.md)

Free Tier ($0/mo)

  • VS Code + Copilot free tier
  • Claude free tier
  • All CLI tools (open source)
  • Self-hosted MCP servers

Power User (~$200-240/mo)

  • Claude Enterprise: $200/mo (500K context)
  • Cursor Pro: $20/mo
  • Premium services: $20-40/mo

📊 Track Your AI Usage

Using multiple AI tools means tracking costs across platforms. The AI Usage Tracker provides unified token usage and cost reporting for Claude Code and Codex CLI.

Quick install:

# pip
pip install ai-usage-tracker
ai-usage-tracker

# npm
npm install -g ai-usage-tracker
ai-usage-tracker-js

Key features:

  • Combined Claude + Codex usage in one view
  • Daily cost breakdown and averages
  • Cache efficiency monitoring (90%+ read rates)
  • Table or JSON output for automation
  • Available as Claude skill: /combined-usage

Typical output: Daily averages, cost per platform, total spend tracking

DAILY AVERAGES (Last 7 complete days)
Claude:      237M tokens/day  |  $123.39/day
Codex:       512M tokens/day  |  $98.76/day
TOTAL:       749M tokens/day  |  $222.15/day

Track actual API usage to optimize spending and identify patterns alongside your subscription costs.

🎓 Who This Is For

You should use this if:

  • ✅ Building production systems with AI
  • ✅ Want terminal-first workflows
  • ✅ Need multi-agent coordination
  • ✅ Value automation over manual work
  • ✅ Cost-conscious but want advanced capabilities
  • ✅ Want evidence-based recommendations

This is NOT for you if:

  • ❌ Prefer GUI-only tools
  • ❌ Want simple autocomplete-only AI
  • ❌ Unwilling to customize/configure
  • ❌ Don't write tests

🤝 Contributing

Contributions welcome for:

  • ✅ Documentation improvements
  • ✅ Setup instruction fixes
  • ✅ Additional evidence/metrics

Not accepted:

  • ❌ "Use X instead" without evidence
  • ❌ Tools not actually used
  • ❌ Promotional content

🔗 Links

Last Updated: 2026-02-12 Analysis Period: Aug 2025 - Feb 2026 (6 months) Total Commits Analyzed: 19,044

About

Evidence-based AI development toolchain recommendations from production usage at WorldArchitect.AI

Resources

Readme

License

MIT license
Activity
Custom properties

Stars

2 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

 
 
 

Contributors