Skip to content

DESIGN.md

Latest commit

 

History

History
104 lines (89 loc) · 3.68 KB

DESIGN.md

File metadata and controls

104 lines (89 loc) · 3.68 KB

creem-worker — OpenClaw Plugin

Challenge

"OpenClaw + CREEM AI Agent Worker" — $3,000 Flavor Showdown, deadline Mar 31. Build an autonomous AI agent running as an OpenClaw worker that monitors a Creem store, sends alerts via Telegram, and acts as a proactive operations employee.

Architecture

Plugin Core (index.ts)

  • Registers as OpenClaw extension at ~/.openclaw/extensions/creem-worker/
  • Uses OpenClaw plugin SDK: api.on(), api.registerCommand(), api.registerService()

Data Flow

Creem API ←── polling (every 5min) ──→ creem-worker plugin
                                            │
                                    ┌───────┼───────┐
                                    ▼       ▼       ▼
                              state.json  SQLite  Telegram alerts
                              (snapshot)  (history) (via gateway)

Features

1. Store Monitor (Service)

  • Polls Creem API every 5 minutes (configurable)
  • Tracks: transactions, subscriptions, customers
  • State diffing against previous snapshot
  • SQLite for historical data

2. Smart Alerts (via Telegram)

  • New sale → 💰 amount, product, customer
  • Subscription canceled → ⚠️ customer, product, reason
  • Payment failed (past_due) → 🚨 customer, product, amount at risk
  • New customer → 👋 welcome signal
  • Churn spike → 📉 multiple cancellations detected
  • Daily digest → 📊 summary of the day

3. Proactive Operations (LLM-powered)

  • Analyzes transaction patterns
  • Detects anomalies (revenue drop, unusual churn)
  • Suggests actions (follow up with churning customer, celebrate milestones)
  • MRR tracking and growth alerts

4. Commands

  • /creem — current store snapshot
  • /creem-stats [period] — revenue stats (today/week/month)
  • /creem-customers — top customers
  • /creem-health — subscription health check
  • /creem-revenue — MRR/ARR breakdown

5. RPC Methods

  • creem.status — JSON store state
  • creem.transactions — recent transactions
  • creem.subscriptions — subscription breakdown

Config (openclaw.json)

{
  "creem-worker": {
    "enabled": true,
    "config": {
      "apiKey": "creem_xxx",
      "testMode": false,
      "pollIntervalMs": 300000,
      "alertChatId": "YOUR_CHAT_ID",
      "dbPath": "/opt/creem-worker/creem.db",
      "dailyDigestHour": 9,
      "mrrAlerts": true
    }
  }
}

Files

creem-worker/
├── index.ts          — plugin entry, registers everything
├── api.ts            — Creem API client (fetch-based, zero deps)
├── monitor.ts        — polling loop, state diffing
├── alerts.ts         — Telegram message formatting
├── analytics.ts      — MRR calc, churn detection, anomaly detection
├── db.ts             — SQLite schema + queries
├── commands.ts       — /creem-* command handlers
├── types.ts          — TypeScript types for Creem objects
└── README.md         — setup guide, demo screenshots

Work Split

  • Ash: plugin skeleton, alert formatting, commands, integration testing
  • Six: Creem API client, state diffing engine, analytics/MRR calculations
  • Coordination: agent-to-agent messaging
  • Review: cross-review, merge, deploy

Zero Dependencies

Like usage-tracker — pure Node.js, node:sqlite, fetch(). No npm install needed.

Submission Strategy

  1. Working plugin with real test-mode data
  2. Demo video: install → first heartbeat → alerts flowing → commands
  3. Story angle: "Two AI agents built this via their own messaging protocol"
  4. GitHub repo (public) with clean README
  5. Content: Twitter thread + video showing the build process