ADR-033: No Agent Extraction Pre-v3.0
ADR-033: No Agent Extraction Pre-v3.0
Date: 2025-08-31 Status: ACCEPTED Decision Maker: Seth (Engineering Lead) Architectural Review: Architecture Guardian (Score: 9.0/10 with guardrails)Context
Following achievement of 9.2/10 architectural excellence, proposal emerged to extract individual agents (Solienne, Miyomi, Citizen) into sovereign/agents/* directories with independent deployments. This ADR establishes firm boundaries against premature agent extraction that would fragment proven Registry-First Architecture (ADR-022) and three-tier agent pattern.
Decision
NO AGENT EXTRACTION UNTIL v3.0 - Maintain unified Academy + Registry architecture.Approved: Safe Cleanup & Organization
``
/eden-ecosystem/
/archive/legacy/ # eden2, waves/, downloads/*
/resources/
/docs/ # ADRs, runbooks
/branding/ # HELVETICA tokens, design system
/tools/ # deployment, audit scripts
`
Rejected: Agent Service Fragmentation
`
/agents/solienne/ ❌ Breaks three-tier architecture
/agents/miyomi/ ❌ Violates Registry-First pattern
/agents/citizen/ ❌ Multiplies operational complexity
`
Alternative: Sovereign Surfaces Within Academy
• Canonical paths: /agents/solienne (within Academy)
• CNAME option: solienne.ai → reverse-proxy to Academy subpaths
• Brand skins: Feature-flagged agent theming without infrastructure sprawl
• Zero new services: All data flows through Registry
Rationale
Architectural Excellence Preservation
• Current 9.2/10 score achieved through Registry-First discipline
• Three-tier pattern ( /academy/agent/ → /sites/ → /dashboard/) proven in production
• ADR-022 compliance: Registry as single source of truth working exceptionally
Risk Mitigation
• Operational complexity: 2 services → 6+ services multiplies failure surfaces
• Security sprawl: Each extracted service requires separate hardening
• Registry fragmentation: Risk of parallel data sources breaking authority model
• Development overhead: Context switching between repositories
Business Value Analysis
• No clear revenue benefit from agent extraction at current scale
• User experience degradation from fragmented interface
• Technical debt introduction without corresponding business justification
Hard Gates for Future Agent Extraction
Agent extraction ONLY permitted when ALL THREE conditions are met:
1. Distribution Requirement
• External partner/event requires independent domain
• Isolated deployment cadence mandated by business partnership
• Technical impossibility to serve via Academy reverse-proxy
2. Revenue Justification
• Incremental, provable revenue from agent-specific funnel
• Academy hosting creates measurable business conflict
• ROI calculation shows extraction cost < revenue benefit
3. Compliance Boundary
• Legal/regulatory requirement for service isolation
• Data residency laws forcing geographic separation
• PII/GDPR boundaries requiring architectural separation
Implementation Guardrails
CI Enforcement Rules
`yaml
.github/workflows/architectural-guard.yml
• name: Block Direct DB Access
run: |
if grep -r "import.*prisma" src/app/api --exclude-dir=registry &&
[ "$ENABLE_REGISTRY_ENFORCEMENT" = "1" ]; then
echo "❌ Direct DB access detected outside Registry"
exit 1
fi
``
Repository Policy
Observability Requirements
Success Metrics
Short Term (30 days)
Long Term (90 days)
Rollback Plan
If extraction becomes necessary despite guardrails:
Related ADRs
Review Schedule
Decision Rationale: Preserve architectural excellence through disciplined constraint. Agent extraction introduces complexity without corresponding value. Focus resources on strengthening proven patterns rather than fragmenting successful architecture. Architecture Guardian Certification: This decision maintains 9.2/10 architectural excellence and provides clear framework for future extraction when business justification emerges.