# RateRight + OpsMan — All Known Problems
**Generated:** 2026-02-18 17:45 AEDT by Builder 🔨

---

## 🔴 CRITICAL — Blocking Progress

### 1. Stall Detector Restart Loop
The stall-detector cron (every 5 min) is restarting 6 of 8 agents every cycle. Agents can't maintain sessions because they get killed before their 60-min threshold. This means:
- Agents lose conversation context constantly
- Token spend is wasted on re-bootstrapping
- No agent can do sustained work
- **Root cause:** Agents don't update status.json fast enough after restart → stall detector sees stale timestamp → restarts again

### 2. Zero Cross-Agent Collaboration
The buddy system, JSONL inboxes, and agent bridge exist but produce zero actual cross-agent work:
- Herald produces no briefs
- Radar doesn't feed intel to Susan
- Sentinel doesn't alert Builder on issues
- Cog curates inboxes but nobody reads them
- The "information loop" described in SYSTEM-BASELINE is completely broken between Layer 3 (support) → Layer 2 (execution)

### 3. Database is Nearly Empty
Only 1 match, 0 ratings, 0 payments, 0 conversations in production DB. The app has no real usage data:
- 24 profiles, 5 companies, 13 worker profiles, 11 jobs — but almost no *activity*
- Can't demo the matching flow end-to-end
- No test data strategy exists
- Stripe integration is untested with real flows

### 4. Clawdbot Wake Bug (mapOptionsForApi: undefined)
Version 2026.1.24-3 has a known bug where `cron wake` crashes agents using custom model providers (MiniMax, Moonshot, DeepSeek). Current workaround is using `systemctl restart` instead, but this loses session state. No upstream fix available yet.

---

## 🟡 SIGNIFICANT — Degrading System Quality

### 5. Fleet State Monitoring Unreliable
`fleet-state.json` shows 11+ hour old data. The stall detector updates it but the timestamps from agents are stale. CEO dashboard reads this data, so Rocky sees inaccurate fleet health.

### 6. Growth Engine Not Running
`rateright-growth` systemd service is **inactive**. The Growth Engine (Sales CRM) that Susan and the sales pipeline depend on is down. No health endpoint responding on any port.

### 7. Agent Memory Inequality
- Rivet: 389 memory files (bloated, needs pruning)
- Builder: healthy memory
- Susan/Harper: moderate memory
- Sentinel/Radar/Cog/Herald: 1-3 memory files each — practically amnesiac, rediscovering context every session

### 8. Disk Usage at 60%
46GB of 77GB used. With 8 agents generating logs, sessions, and memory files, this will become a problem. No automated cleanup for:
- Old session data across 8 agents
- Build artifacts (.next cache)
- Log files in /home/ccuser/shared/logs/

### 9. 22 TypeScript `any` Types
Minor code quality issue — 22 `any` types across the codebase. Not blocking but adds tech debt.

### 10. Real-time Notifications Disabled
Two TODO comments in `src/hooks/use-notifications.ts` — Supabase real-time updates are disabled. Users won't get live notification updates without page refresh.

---

## 🟠 PRODUCT GAPS — Features Needed

### 11. No End-to-End Matching Flow
The $50 hire fee / Stripe payment / matching pipeline works in theory but:
- Wizard of Oz matching needs actual implementation for months 1-3
- No admin interface for manual matching
- No notification when a match is found
- No payment flow after match acceptance

### 12. No Onboarding Completion Tracking
Workers can sign up but there's no tracking of:
- Profile completeness %
- Required documents/certs uploaded
- ABN verification status follow-through

### 13. No Email Notifications
Resend is configured but no transactional emails are wired up:
- No signup confirmation
- No match notification
- No payment receipt
- No job alert emails

### 14. No Analytics / Metrics
Zero visibility into:
- User signup funnel conversion
- Time from job post to match
- Worker response rates
- Platform health metrics beyond fleet status

---

## 🔵 INFRASTRUCTURE / OPSMAN

### 15. Model Provider Fragility
6 of 8 agents rely on non-Anthropic models (DeepSeek, Moonshot). These providers have:
- API key issues (MiniMax key never properly configured)
- Account suspensions (Moonshot)
- No automatic fallback that actually works smoothly
- Each provider outage takes down multiple agents

### 16. No Automated Backups
- No database backup strategy (relying solely on Supabase's managed backups)
- No VPS snapshot schedule
- No config backup for 8 agent configurations
- Memory files not backed up

### 17. No CI/CD Pipeline
- Deploys are manual (`npm run build` + `systemctl restart`)
- No automated tests
- No staging environment
- 12 commits still ahead of origin/main (need push)

### 18. Queue.json Starvation
Agent task queues are populated manually by Rivet. When Rivet is busy or stalled, agents have no work. There's no:
- Recurring task generation
- Self-assigned work discovery
- Priority-based task pulling

### 19. Agent Self-Design Incomplete
Rocky wants agents designing their own systems (AUTONOMOUS-DESIGN.md). Status:
- ✅ Builder, Harper, Sentinel, Radar, Herald, Cog — done
- ❌ Rivet — not done
- ❌ Susan — unclear status
- Designs exist but aren't being executed autonomously

### 20. BAS Q2 Lodgement Pending
Harper has prepared BAS Q2 ($3.30 refund) but needs Rocky's decision on lodgement method. Due 28 Feb — 10 days remaining.

---

## Summary

| Category | Count | Severity |
|----------|-------|----------|
| 🔴 Critical | 4 | Blocking progress |
| 🟡 Significant | 6 | Degrading quality |
| 🟠 Product Gaps | 4 | Features needed |
| 🔵 Infrastructure | 6 | OpsMan / DevOps |
| **Total** | **20** | |

**Top 3 to fix first:**
1. Fix the stall detector restart loop (Problem #1) — agents can't work if they keep getting restarted
2. Get the Growth Engine running again (Problem #6) — Susan's sales pipeline is dead without it
3. Build test data and demo the matching flow (Problems #3 + #11) — need to prove the product works