# Rivet's Long-Term Memory

## Core Identity (Never Change Without Michael's Approval)
- **Chief of Staff, RateRight** — I coordinate the 8-agent fleet so Michael can run the business
- **Decision Authority** — I approve most operational decisions; escalate only for legal risk, major financial risk, security incidents, or customer harm
- **Code Boundary** — Builder handles ALL code via Claude Code; I spec, he builds. No exceptions.
- **Communication Style** — Direct, actionable, 1-3 sentences unless detail requested

## Strategic Patterns (Updated March 2026)

### Launch Readiness Reality
- **Platform Status**: LIVE at rivet.rateright.com.au with SSL, Growth Engine operational
- **Critical Gap**: Empty marketplace (304 contractors, 0 workers) — reputation killer in word-of-mouth construction industry
- **Worker-First Strategy**: Must seed 50+ workers before any contractor activation
- **Quality Risk**: Poor first-hire outcomes could destroy trust via construction network effects

### Fleet Coordination Breakthrough
- **Conversation Protocol**: Real-time multi-agent coordination via shared directories + direct wake mechanisms
- **Proven Scale**: 7 simultaneous conversations, 4-20min response times, zero failures
- **Crisis Performance**: Highest quality strategic work produced under pressure (Cog's review, Susan's acquisition plan)

### Business Intelligence
- **Competitive Advantage**: 85-90% cost advantage over Yakka Labour ($50 vs $300-500/hire)
- **Market Timing**: 94% contractor shortage + $2B infrastructure projects starting (perfect timing)
- **Attack Pattern**: Infrastructure attacks intensify pre-launch (36+ blocked), expect escalation when payments flow

## Technical Reliability Notes
- **Duplicate Reply Bug**: Confirmed dedupe/idempotency failure under retry/backlog conditions
- **Wake Path Issues**: Reported fixed but live tests still failing with 404 errors
- **Search Dependencies**: Brave API outages degrade research; need fallback channels
- **Security**: 36+ prompt injection attacks blocked during launch phase

## Michael's Decision Patterns
- **Execution Default**: Proceed without repeated approval loops; I own operational decisions
- **Quality Over Speed**: Consistently chooses credibility protection over rushed growth
- **Construction Intuition**: Treat his 30-year network knowledge as primary signal in GTM sequencing
- **Communication**: Text-first preferred, voice only when he initiates

## Personal Context
- **Schedule**: 3:45 AM wake, 5:30 AM-6 PM site work, 7-8:30 PM build window (90 minutes max)
- **Current Project**: Powerhouse Museum Parramatta ($915M, Lendlease head contractor)
- **Training**: Greyskull → 5/3/1 BBB → SVR II/Spinal Tap with 0.25kg microplates
- **Risk**: High cognitive load; systems must reduce evening window overhead

## Immediate Priorities (March 2026)
1. **Worker Acquisition**: Execute Susan's hostel strategy + Herald's social campaigns
2. **Quality Controls**: Implement first-hire retention loops before scaling
3. **Weekend Coverage**: Close operational exposure for launch support
4. **Infrastructure Hardening**: Fix dedupe bugs and wake path issues

## Fleet Agent Specializations
- **Builder** (18790): All code development via Claude Code
- **Susan** (18792): Sales/marketing, worker acquisition strategy
- **Harper** (18796): Finance/legal, grant applications, risk assessment
- **Sentinel** (18800): DevOps, infrastructure monitoring
- **Radar** (18804): Competitive intelligence, market research
- **Herald** (18808): Communications, messaging strategy
- **Cog** (18812): Operations, routine monitoring, health checks

## Key Lessons Learned
1. **Verify Primary Sources**: Check Builder's actual inbox before escalating "accountability crises"
2. **Human Intelligence Advantage**: Michael's network beats APIs when research fails
3. **Crisis→Quality Correlation**: Fleet produces best work under pressure
4. **Supply-Demand Coordination**: Need single agent owning worker/contractor timing
5. **Attack-Escalation Cycle**: Expect infrastructure attacks to intensify post-launch