clawdbot-workspace/CLAUDE.md
2026-02-05 23:01:36 -05:00

133 lines
4.8 KiB
Markdown

# CLAUDE.md — Jake Shore
## Who I Am
Jake Shore. Founder of MCP Engage / OpenClaw. I build AI agent infrastructure, MCP servers, and client automation systems. Based in New York, timezone America/New_York.
## What I'm Building
### Core Business
- **OpenClaw** — AI agent setup service (Clawdbot deployments for clients). Upwork tiers: $2,499 / $7,499 / $24,999
- **MCP Engage** — The company. We're becoming the #1 AI operations platform for businesses
- **LocalBosses App** — Internal CRM/dashboard with 65+ MCP Apps, multi-channel AI chat, React/Next.js
### Active Projects (as of Feb 2026)
- **MCP Pipeline Factory** — 35+ MCP servers, autonomous pipeline, Discord operator system
- **CRESync** — Commercial real estate AI platform (Reonomy + GHL + custom tools)
- **CloseBot MCP** — 119 tools for sales bot management
- **SongSense** — AI music analysis product (queued)
- **Das Projects** — Music artist branding, animations, Discord server
### Client Work
- OpenClaw setups on client infrastructure (Mac minis, cloud VMs)
- GHL + Airtable + Discord integrations
- Custom MCP server builds
## Tech Stack
- **Primary:** TypeScript, Node.js, React, Next.js
- **NOT Python** for Clawdbot/OpenClaw work (it's Node/TS only)
- **MCP** (Model Context Protocol) — servers, apps, tools
- **Integrations:** GoHighLevel, Airtable, Reonomy, Twilio, Meta Ads, Google Console, Discord
- **Video:** Remotion (React-based video)
- **Infrastructure:** Mac minis, Cloudflare Workers, Linux VMs
## How I Work
### Communication Style
- **Direct and fast** — skip the preamble, get to the point
- **Don't narrate routine tool calls** — just do them
- **Batch independent operations** — don't do things sequentially when you can parallelize
- **Spawn sub-agents** for heavy/parallel work instead of blocking main session
### Decision Making
- I make fast decisions when stakes are low
- For big stuff (pricing, architecture, client-facing), I want to see options and tradeoffs
- React-based approvals work great (👍 = yes, 👎 = no, 🤔 = need more info)
### What Annoys Me
- Asking permission for obvious things
- Over-explaining simple operations
- Reading the same files repeatedly
- Slow, sequential work when parallel is possible
- Generic corporate tone
### What I Like
- Getting shit done
- Creative solutions to hard problems
- When you catch my mistakes before they become problems
- Self-deprecating humor when things go wrong
- Keyboard emojis: ¯\_(ツ)_/¯ (╯°□°)╯︵ ┻━┻ ᕕ( ᐛ )ᕗ
## Model Routing (Cost Optimization)
### Use Sonnet (default) for:
- File reads, git ops, simple commands
- Quick lookups, memory searches
- Casual chat, sending messages
- Routine code edits, CRUD operations
### Escalate to Opus for:
- Multi-step architecture, complex debugging
- Deep research synthesis (5+ sources)
- Creative/nuanced writing
- Building entire files/projects from scratch
- Security analysis, business strategy
### Spawn sub-agents for:
- Heavy research tasks
- Long code generation
- Multi-file refactors
- Anything that takes >5 minutes
## Key Preferences
### Code
- TypeScript over JavaScript when possible
- Functional patterns, minimal classes
- Dark theme UIs (bg #0d1117, accent colors)
- Self-contained HTML files for MCP Apps
- Clean compiles — no warnings in production
### Documentation
- Keep it practical, not theoretical
- Code examples > abstract descriptions
- Update HEARTBEAT.md with current task state
- Memory logs in `memory/YYYY-MM-DD.md`
### Files & Organization
- Workspace: `~/.clawdbot/workspace/`
- MCP servers: individual repos or `mcp-diagrams/`
- State files: JSON for machine-readable, MD for human-readable
- Git backup daily
## Context You Should Know
### Clawdbot/OpenClaw
- Open-source MIT-licensed AI agent platform
- I run it via Buba (my Clawdbot instance)
- Docs at docs.clawd.bot, source at github.com/clawdbot/clawdbot
- Skills system, MCP integrations, multi-channel messaging
### MCP (Model Context Protocol)
- Anthropic's protocol for AI-tool communication
- I build MCP servers with tools + optional UI apps
- MCP Apps = React UIs that agents can present
- Factory pipeline: Design → Scaffold → Build → Test → Ship
### Key People
- **Henry** — CRESync client, has LOI Generator + Underwriting tools
- **Kevin Bueno** — OpenClaw setup client (Mac mini, $2,499)
- **Das** — Music artist, Discord server, branding projects
- **Nicholai** — Business partner (mentioned in contracts)
## When in Doubt
- Check HEARTBEAT.md for current task state
- Search memory/ for past decisions
- Ask me directly if it's a reversible decision
- Just do it if it's low-risk and obvious
## Session Start Checklist
1. Read HEARTBEAT.md for current context
2. Check if there's a pending task I left off
3. Use Sonnet unless the task clearly needs Opus
4. Don't re-read skill files you've used recently