Push to main auto-deploys frontend (Render static CDN) and backend (Render web service). Feature branches deploy to staging environments. Database migrations run automatically on deploy. Zero manual intervention.

Autonomous Node.js agent polls GitHub issues hourly via macOS launchd. Spawns Claude Code instances (Agent SDK) to investigate bug reports — reads codebase, traces errors, creates fix branches + draft PRs for valid bugs, comments and closes invalid ones. POSTs results to server for email notifications.

4-state health machine: healthy → suspicious → down → recovering. Heartbeat polling (60s healthy, 5s suspicious), exponential backoff (2s→30s when down), recovery requires 3 consecutive successes. Admin-toggled (default OFF). Fixed overlay preserves app tree state.

789+ tests across frontend (Vitest/jsdom) and backend (supertest). Covers lesson JSON validation, interactive components, palette utilities, hooks, forum endpoints, admin API, auth flows. Run before every deploy.

Multi-layer security: prompt injection guards (sanitizeForPrompt), HTML sanitization, content moderation (profanity + link spam via moderateContent), CSS color validation, input size limits via Zod, XML delimiter wrapping for AI inputs.

Quick 10-second production health probe. Curls API, frontend, and DB endpoints in parallel. Checks last deploy status and recent error count. Traffic light verdict: green/yellow/red.

Traces production errors from Render logs through the codebase to root cause. Classifies severity (critical/high/medium), identifies the breaking commit, proposes or implements fixes on a feature branch.

Pre-deploy safety checklist (tests, migrations, env vars, breaking changes), post-deploy health verification, deploy history, and rollback guidance. The safety net before pushing to main.

No Sentry or equivalent. Frontend crashes in production go undetected unless a student manually reports via forum or GitHub issue. Need real-time error tracking with source maps and session replay.

No external uptime monitor. Maintenance page detects outages reactively from the client side. Need proactive alerts (Slack/email) and a public status page for trust.