Most people use Claude Code to write code. Developer Edwin Trebels is using it to run his entire go-to-market operation. His new open-source project, Claude Code Growth OS (MIT licensed), turns the Anthropic AI agent into a complete operating environment for marketing, sales, product, and retention — all tracked in git with full history, diffs, and blame views.

The Core Architecture

The system is built around three layers: hooks that enforce guardrails across sessions, commands that invoke daily rituals by name (/morning-briefing, /end-of-day, /weekly-review), and skill templates spanning all four growth functions. Everything lives in markdown files inside the repo — your pipeline goes in ops/pipeline.md, customer accounts in ops/customers.md, and a cross-function feedback loop in ops/feedback-log.md. The setup includes five Claude Code hooks: session-start context injection, state re-injection before compaction (so long sessions don't lose thread), a pre-commit secret-guard that blocks API keys and tokens from slipping into git, sensitive-file protection for .env and credentials, and an end-of-day nudge to log what shipped.

Daily Rituals and the Bowtie Model

The system models go-to-market as a bowtie: marketing and sales on the left side (acquisition), product and retention on the right (post-sale adoption and renewal). The morning briefing reads your priorities.md, yesterday's daily-log.md, and pipeline.md to surface today's top three actions. Skills work on both sides of that bowtie — lead-qualify a new opportunity on day one; meeting-prep before a call; onboarding-handoff when a deal closes (carrying the value hypothesis across the seam); account-health scoring at day 60 to start renewal motion; product-signal routing retention and field signals to the roadmap. Two feedback loops connect the halves: a MARKETING-ACTION line for sales→marketing (when an objection surfaces three times in a month, flag it), and a RETENTION-RISK line for post-sale→product (usage drops two weeks, only one of three workflows adopted triggers an alert).

Hooks That Earn Their Place

The pre-compact hook re-injects priorities and the latest log when Claude Code compacts context — your state lives in files, so the window can't lose the thread. The pre-commit-guard blocks commits if staged changes match patterns for API keys, private keys, or tokens. Protect-files.sh stops the agent from writing to .env and credential stores entirely. Session-start opens every day with priorities and fresh cross-function signals; stop-reminder closes it. There's also a web-bootstrap hook that runs only in remote/Claude-Code-on-the-web sessions (when CLAUDE_CODE_REMOTE=true) and installs shellcheck so you can lint hooks in-session, matching CI — skipped on your local machine, idempotent, never blocking.

The 30-Second Demo

Clone the repo, open it in Claude Code, and run /demo-briefing. It executes the whole morning ritual against fictional sample data in demo/ — surfacing three priorities for today (including a slipping deal at Rheinkraft Manufacturing with eight days of silence), tasks pulled from yesterday's meetings, and flagged signals for both feedback loops. Nothing is real; delete demo/ whenever you like. The fictional pipeline shows six deals, one slipping, two with no next step, and an onboarding account at risk — enough context to see the full motion without exposing sensitive data.

Open Core Model

The repo itself is MIT-licensed and free: hooks, daily rituals, go-to-market skill templates across all four functions, plus the 30-second demo. The paid layer (via langoptima.com/features/growth) adds wider automation via n8n for vendor management, ICP checks, pre- and post-meeting briefings, follow-ups, and multilingual content processes through Blackbird.io — plus Edwin's own judgment filling in positioning, pricing, and playbooks. That part took twenty-two years and doesn't come in a folder.

Portability Note

The .claude/skills/ and .claude/commands/ directories are plain markdown on the Agent Skills spec, so they port to Cursor, Windsurf, and Codex with minimal changes. The hooks/ and settings.json are Claude-Code-specific by design — Edwin notes that's fine; the rituals and skills are the part worth taking elsewhere.

Key Takeaways

  • Your entire go-to-market motion (marketing, sales, product, retention) now lives in git with full history and diffs
  • Five Claude Code hooks handle session continuity, secret guarding, and file protection across contexts
  • Daily rituals (/morning-briefing, /end-of-day, etc.) are just markdown prompts you can copy and customize
  • The bowtie model connects acquisition to retention through two explicit feedback loops
  • MIT licensed with a 30-second demo — no API keys or MCP required out of the box

The Bottom Line

This is what happens when hackers get tired of their sales process living in someone's head. Git-tracked playbooks, hooks that survive context compaction, and a morning briefing that reads your pipeline — if you've been treating Claude Code as just an autocomplete engine for code, Growth OS shows you how much more runway it has.