chief-of-staff

Claude Home Base

An always-on AI cofounder running on your Mac. DM it in Slack, it responds with full access to your codebase, tools, and context. $200/month flat. You own the whole stack.

What this is

A complete setup for turning a spare Mac (Mini, MacBook Air, whatever) into a dedicated AI server:

• Slack bot that wraps Claude Code's CLI — DM it or @mention it in channels
• Cloudflare Tunnel for production-grade Slack integration (HTTP Events API, not Socket Mode)
• Plugin marketplace with skills for creative direction, coding, image generation, brainstorming, and more
• Identity system — your AI writes its own personality, keeps a diary, compounds over time
• Setup guide — step-by-step, with dark mode, interactive checklists, and concept explainers

What you need

• A Mac you can leave running (Mac Mini, old MacBook Air, etc.)
• Claude Code Max subscription ($200/month)
• A Slack workspace
• A domain for Cloudflare Tunnel (any domain works)

Quick start

1. Follow the setup guide at nityeshaga.github.io/claude-home-base — it walks you through everything step by step
2. Set up hardware — plug in your Mac, configure it for always-on use
3. Deploy the Slack bot — Cloudflare Tunnel + Flask, production-standard
4. Install the starter kit — the final step in the guide has you paste one prompt into Claude Code. It clones this repo, installs plugins, asks you a few questions, and writes its own identity. You watch it come alive.

What's in the box

bot.py                  # Slack bot (Flask + HTTP Events API)
index.html              # Setup guide (GitHub Pages)
CLAUDE.md.example       # Template for your AI's operations manual
identity.md             # Your AI's soul (principles + self-authored identity)
about-you-and-how-you-came-to-life.md  # Origin story template
.env.example            # Configuration template
requirements.txt        # Python dependencies

plugins/
├── coding/             # Precision coding tools
│   └── skills/
│       └── make-precise-ui/  # Pixel-perfect UI from Figma designs
├── creative/           # Creative direction, writing, brainstorming
│   └── skills/
│       ├── creative-lead/    # Creative direction for any project
│       ├── lets-brainstorm/  # Timed coaching sessions
│       ├── help-me-write/    # Collaborative writing (keeps your voice)
│       └── interview-me/     # Timed discovery interviews
├── more-ai/            # Gemini and OpenAI image generation, thinking
│   └── skills/
│       ├── gemini-imagegen/
│       ├── openai-imagegen/
│       └── gemini-thinking/
└── experimental/       # Operational workflows, debiasing, prompt engineering
    └── skills/
        ├── briefing/         # Email, briefs, action tracking
        ├── are-you-sure/     # Blind debiasing for claims and opinions
        ├── prompt-engineer/  # AI prompt writing and review
        └── investigate-yourself/  # Forensic self-diagnosis

Architecture

You (anywhere) → Slack → Cloudflare Tunnel → Your Mac → Claude Code CLI
                                                          ↓
                                              CLAUDE.md + identity.md
                                              + plugins + skills
                                              + full filesystem access

Bot features

• HTTP Events API via Flask — production-standard, stateless
• Async processing — responds to Slack within 3 seconds, runs Claude in background
• Agentic channel behavior — decides when to respond, stays silent when not relevant (SKIP)
• Thread continuity — session IDs persist per thread
• File handling — downloads attachments, auto-uploads files mentioned in responses
• Proactive messaging — send DMs, post to channels, reply in threads via CLI
• Streaming output — real-time responses as Claude generates
• Native tables — markdown tables in responses render as real Slack tables (Block Kit markdown block)
• Interactive buttons — button clicks and menu picks route back into the thread's Claude session as messages, so your AI can offer approve/hold/snooze choices and act on the answer (requires Interactivity enabled in your Slack app config; Request URL = the same /slack/events endpoint)
• In-thread stop — type a bare stop in a thread where the bot is mid-run to interrupt it (like Esc in the terminal); the session survives with full context, so your next message steers it in the new direction
• Mid-turn steering — message a thread while the bot is mid-run and it sees your message at the next tool-call boundary, inside the same turn (like typing without Esc in the terminal); no more waiting for the whole task to finish before you can course-correct

License

MIT