Giving away the exact Claude Code framework I'd use to run all of GTM from one terminal: ads, outbound, RevOps, content. Free.
Without real structure, Claude drifts after a couple of sessions and the output falls apart. With it, the system gets sharper every week. Our head of growth Ivan Falco built the whole thing.
It comes down to 5 layers, each with one job:
→ Identity: a single CLAUDE.md that loads every session. It defines who the agent is, the rules it follows, and how it routes each task.
→ Rules: they live in a rules folder and load automatically based on where you're working. Outbound rules in outbound, ad rules in ads.
→ Skills: step-by-step workflows that only activate when a task matches them, like building a sequence, auditing a campaign, or writing a report.
→ Agents: subagents built for one job each, like research, QA, or code review. Each runs in its own context so it doesn't clog the main thread.
→ Memory: a MEMORY.md that persists across sessions. What the agent learns once carries into every session after.
The layer that makes it durable is governance. You split the brain from the muscle.
The brain handles governance, the muscle handles execution. An execution agent never rewrites its own rules. When it spots something worth changing, it files a proposal. You review it, confirm it, and only then does the change go live and get logged.
That single rule is why the system compounds instead of drifting.
Ivan put the whole framework into one guide: the folder structure, the governance loop, every skill and agent.
Reply "FRAMEWORK" and I'll send it over. Must be following so I can DM.