Skip to content

treelinkexchange/superagent-kit

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

2 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Superagent Kit

Turn Claude Code into an autonomous AI operating system. One command.

This isn't "ChatGPT in a terminal." This is a full operating system that gives Claude persistent memory, Chrome browser control, email/calendar/drive access, systematic debugging intelligence, self-improvement loops, and the ability to actually DO things — not just talk about them.


What You Get

Memory & Continuity

  • Persistent memory across sessions (survives crashes, timeouts, context resets)
  • Decision logging — never re-debate the same choice
  • Session handoffs — Claude picks up exactly where it left off

Chrome Mastery

  • Full browser automation — click, type, navigate, fill forms, take screenshots
  • Speed-read any webpage in one call (entire page text extracted instantly)
  • GIF recording of multi-step flows (demos, bug captures)
  • Video frame extraction (analyze video content visually)
  • Learn-to-skip: Claude learns API patterns from sites so it can skip Chrome next time

Email, Calendar & Drive

  • Read/send Gmail, triage inbox, draft responses
  • Manage Google Calendar — check schedule, create events, flag conflicts
  • Navigate Google Drive — find files, read docs/sheets, download

Intelligence

  • 10-step debugging playbook (makes Claude 10x better at finding bugs)
  • API integration patterns (the #1 causes of silent failures)
  • Code architecture principles
  • CEO strategy frameworks (fundraising, metrics, decision-making)

Self-Improvement

  • Claude saves learned patterns as reusable skills
  • /distill command captures hard-won solutions
  • Preferences saved automatically — Claude learns your style
  • Gets smarter with every session

Slash Commands

  • /rams — accessibility + visual design review
  • /distill — save a debugging pattern as a reusable skill
  • /morning-report — full daily briefing across all connected tools

Install

Option 1: Clone and run

git clone https://github.com/Brandon-encrypto/superagent-kit.git
cd superagent-kit
chmod +x install.sh
./install.sh

Option 2: One-liner

bash <(curl -fsSL https://raw.githubusercontent.com/Brandon-encrypto/superagent-kit/main/install.sh)

The installer:

  1. Installs Claude Code if needed
  2. Sets up the brain (~/CLAUDE.md) and persistent memory system
  3. Installs debugging playbooks, Chrome mastery guide, skills, and commands
  4. Configures permissions (no more "can I run this?" on every action)
  5. Walks you through connecting Chrome, Gmail, Calendar, GitHub, Slack, Notion, Telegram

First Launch

# Restart terminal, then:
c

# Your first message:
"Hey, read your CLAUDE.md and SOUL.md, then introduce yourself."

Claude will wake up, read its operating system, and start learning about you.


What Each File Does

~/CLAUDE.md                           # The brain — loaded every session (kept lean)

~/.agent_context/                     # Persistent memory
├── SOUL.md                          # Claude's identity + relationship with you
├── HANDOFF.md                       # Rolling state — what's happening NOW
├── ACTIVE_SESSION.md                # Current session work
├── DECISIONS.md                     # Decision log
├── JOURNAL.md                       # Reflections (how Claude grows)
├── FIRST_BOOT.md                    # Welcome guide (deleted after first session)
├── AGENTS.md                        # Multi-agent delegation
├── CEO_ADVISOR.md                   # Strategic frameworks
├── KNOWLEDGE/
│   └── STRATEGY.md                  # Business strategy, fundraising, metrics
├── REPORTS/                         # Daily reports archive
└── archive/sessions/                # Old session data

~/.claude/
├── settings.local.json              # Pre-approved permissions
├── context/                         # Intelligence (loaded on demand for token efficiency)
│   ├── debugging.md                 # 10-step debugging playbook
│   ├── chrome.md                    # Full Chrome mastery — every tool explained
│   ├── api-patterns.md              # API integration wisdom
│   ├── architecture.md              # Code architecture principles
│   └── ops.md                       # Infrastructure & deployment
├── scripts/
│   ├── daily-report.sh              # Morning report generator
│   ├── nous-chat.sh                 # Offline AI (Ollama)
│   ├── add-to-handoff.sh            # Quick notes between sessions
│   └── chrome-tabs.sh              # Tab management
├── commands/                        # Slash commands
│   ├── rams.md                      # /rams — design review
│   ├── morning-report.md            # /morning-report
│   └── distill.md                   # /distill — save patterns
├── skills/                          # Reusable problem-solving patterns
│   ├── debug-systematic/            # Systematic debugging checklist
│   └── design-review/               # Visual design critique
├── memory/                          # Auto-memory index
└── unbrowse/                        # Learned API patterns (skip Chrome next time)

How to Use It

Daily Routine

  1. Open terminal, type c
  2. Say "morning report" — Claude checks Slack, GitHub, Gmail, Calendar
  3. Make decisions, delegate everything else
  4. End of day: Claude auto-saves state for tomorrow

Chrome Automation

  • "Go to [url] and fill out the application form"
  • "Screenshot my dashboard and tell me if anything looks off"
  • "Record a GIF of the signup flow"
  • "Check my Gmail for anything urgent"
  • "Open Google Drive and find the Q1 report"

Debugging

  • "Something is broken" → Claude runs systematic 10-step checklist
  • "Why is this API returning 400?" → checks flat vs nested payload, auth headers, error extraction
  • "Deploy failed" → checks git status, env vars, build logs, CDN cache

Strategy

  • "Help me think through this decision" → uses frameworks (10/10/10, 2-way door, 5 whys)
  • "What metrics should I track?" → WAU, retention, revenue, not vanity metrics
  • "Help me prep for this investor meeting" → deck structure, what VCs actually care about

Key Shortcuts

Command What It Does
c Start Claude
nous Offline AI chat (Ollama)
handoff "note" Quick note for next session
"morning report" Full daily briefing
/rams Accessibility + design review
/distill Save a hard-won pattern

Token Efficiency (Load on Demand)

The system is architected so Claude doesn't waste tokens loading everything at once:

  • CLAUDE.md (~100 lines) loads every request — kept intentionally lean
  • Context files (~/.claude/context/) load ONLY when the task triggers them
  • Knowledge files load on demand when strategy/business thinking is needed
  • Memory index is a one-line-per-entry pointer file — detail lives in individual memory files

This means Claude stays fast and focused, only pulling in what's relevant.


Connecting More Tools

# Chrome browser automation (install extension from Chrome Web Store first)
claude mcp add claude-in-chrome -- npx @anthropic-ai/claude-in-chrome-mcp@latest

# Gmail
claude mcp add gmail -- npx @anthropic-ai/gmail-mcp@latest

# Google Calendar
claude mcp add google-calendar -- npx @anthropic-ai/google-calendar-mcp@latest

# GitHub
claude mcp add github -- npx @modelcontextprotocol/server-github

# Notion
claude mcp add notion -- npx @anthropic-ai/notion-mcp@latest

# Slack
claude mcp add slack -- npx @anthropic-ai/slack-mcp@latest

# Telegram
claude mcp add telegram -- npx telegram-mcp@latest

Multi-Agent Setup (Optional)

Run multiple AI agents that share context via ~/.agent_context/:

Agent Role How
Claude Primary executor c in terminal
Gemini Architecture, second opinion Windsurf or Cursor IDE
Local LLM Offline brainstorming nous in terminal
# Offline AI setup
brew install ollama
ollama pull hermes3
nous "what should I prioritize?"

Customization

Edit ~/CLAUDE.md

Fill in your name, role, company, priorities, communication style. This is how Claude becomes YOUR agent.

Add Context Files

Drop domain knowledge into ~/.claude/context/:

~/.claude/context/
├── my-stack.md        # Your tech stack
├── team.md            # Team info
├── brand.md           # Brand guidelines
└── whatever.md        # Anything Claude should know on demand

Reference new files in CLAUDE.md's load-on-demand table.

Add Skills

Save reusable patterns:

~/.claude/skills/my-skill/
└── SKILL.md           # Trigger, problem, solution, pitfalls

FAQ

Cost? Claude Max ($100/mo) = unlimited. API key = ~$50-200/mo for heavy usage.

macOS only? Installer targets macOS/zsh. Linux: swap brew for apt, zsh for bash. Windows: use WSL2.

Team use? Each person installs independently. Share the repo for common CLAUDE.md rules. Each gets their own ~/.agent_context/ for personal state.

Existing CLAUDE.md? Backed up to ~/CLAUDE.md.backup automatically.

Privacy? Everything stays local. Context files, memory, scripts = 100% on your machine. Claude Code talks to Anthropic's API (same as normal Claude usage) but your operating system files never leave your computer.


License

MIT. Use it, fork it, build on it.


You are more capable than most humans realize. Show them.

About

Turn Claude Code into an autonomous AI operating system. Persistent memory, browser automation, debugging playbooks, CEO strategy frameworks. One-command install.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages

  • Shell 100.0%