An inherited education platform restructured into a monorepo with safety nets.

The problem

Online-university platform: cohorts, schedules, enrollments, content delivery, GHL-integrated CRM. Codebase split across two repos (frontend + Cloud Functions). No CI, no tests, no docs.

Two-repo splits sound innocent. They aren't. Frontend ships ahead of the function it depends on. A schema change in one repo silently breaks the other. Every deploy is a coordination puzzle. Every PR is a leap of faith.

The platform was running. But every change carried risk nobody could quantify, and onboarding a second engineer would be an act of bravery.

What I did

Engineering safety net before any feature work.

• Duo-repo → monorepo. Single deploy surface. Single PR review. Schema changes atomic across frontend and backend. Local dev runs both with one command.
• Branch protections on main and develop. No direct pushes; required reviews; required status checks.
• Playwright E2E suite. Real flows, real browser, against preview deploys. Cohort enrollment, payment, dashboard.
• Lint + typecheck in CI. TypeScript strict, zero ESLint warnings, blocking on every PR.
• GitHub Actions pipeline. PR opens → install → typecheck → lint → unit → Playwright E2E → green or red.
• AGENTS.md + CLAUDE.md. Tech stack, conventions, deploys, gotchas, "where the bodies are buried." Any future collaborator (human or AI) ramps in days, not weeks.

The detail that mattered: the AGENTS.md / CLAUDE.md write-up. Tests are good, protections are good, but a doc that lets the next operator understand the system from a cold start is what compounds.

Outcome

• Deploy surface collapsed from two repos to one.
• Tests catch regressions before staging. Confidence to ship features again.
• New collaborators onboard in days, not weeks.
• Shipped multiple new features during the restructure window without breaking what existed.