Stats
Links
Categories
Mentilead GrowthOS
An AI operating system for running a Shopify app portfolio. GrowthOS puts an autonomous COO agent in charge of daily operations — reading state across marketing, publishing, experiments, and development, delegating to 18 specialized skills, and coordinating code work across projects via a cross-agent comms protocol. The human holds direction and values. The agents operate.
This isn't a marketing template. It's the infrastructure for a live experiment: Can one person replace every operational function with AI agents?
How It Works
The COO agent runs a daily briefing cycle:
- Reads all state — marketing metrics, experiment observations, publishing cadence, development roadmap, and pending agent-comms
- Identifies what matters — overdue follow-ups, anomalous metrics, stale experiments, unprocessed Claude Code results
- Acts autonomously on routine work — research, state updates, delegations to other skills, writing instructions for Claude Code
- Escalates only decisions requiring founder judgment — publishing, outreach to real people, strategy changes, spending
- Presents a briefing with completed actions, alerts, and ranked recommendations
- Routes approved work directly to the right skill with full context
Everything persists in local markdown files. No APIs, no databases, no accounts. Human-readable and git-trackable.
The Experiment
GrowthOS tracks one core question: Can one person replace every operational function with AI agents until the only irreplaceable role is holding direction and values?
This runs on real revenue, real customers, and real consequences.
The primary KPI is the Autonomy Score — the percentage of operational decisions that required no human input. As of Chapter 1, it's around 50%. The goal is to push it higher by identifying which decisions genuinely need a human and which are just habit.
Observations get logged as they happen — agent decisions, failures, escalations, contradictions with enterprise AI narratives. These observations feed into published content via a five-pass quality pipeline.
Skills
Operations
| Skill | What It Does |
|---|---|
| COO | Autonomous operating officer — reads all domains, delegates, executes routine work, escalates the rest |
| Weekly Review | Structured review ritual with metrics synthesis, experiment check-ins, and priority planning |
| Status | Quick marketing health dashboard with metrics and benchmarks |
| Next | Recommends the single highest-impact marketing action |
Marketing
| Skill | What It Does |
|---|---|
| Positioning | ICP, Jobs-to-be-Done, and positioning using April Dunford's framework |
| ASO Optimizer | App Store listing optimization (title, subtitle, keywords, description, visuals) |
| Content Strategy | Content planning with idea generation, calendar, distribution, and briefs |
| Outreach | Cold email campaigns with segment targeting and template generation |
| Partnerships | Complementary app discovery, agency outreach, and pipeline tracking |
| Review Manager | Review acquisition strategy with Shopify policy compliance |
| Metrics Dashboard | Funnel tracking against Shopify benchmarks |
| Experiment Engine | ICE-scored experiment management with hypothesis tracking |
Publishing
| Skill | What It Does |
|---|---|
| Publisher | Drafts Substack, LinkedIn, and Medium posts from observations with Personal Voice Protocol and anti-AI quality gate |
| Observation Logger | Captures agent decisions, failures, escalations, and contradictions |
| Signal Monitor | Tracks enterprise AI signals and surfaces contradictions with experiment findings |
| Autonomy Tracker | Tracks the Autonomy Score — the experiment's primary KPI |
Infrastructure
| Skill | What It Does |
|---|---|
| Init | Sets up the marketing workspace for single or multi-app portfolios |
| Idea Validator | Validates app ideas with a 5-Signal Framework before building |
Agent-Comms Protocol
GrowthOS runs across two agents: Cowork (Claude Desktop, where the COO lives) and Claude Code (terminal, where code changes happen). They coordinate via a mailbox system:
agent-comms/
outbox/ # Instructions written by Cowork, run in Claude Code
inbox/ # Results written by Claude Code after completing work
archive/ # Completed exchanges
project-registry/ # One file per project with architecture and gotchas