TorbatYar/docs/architecture/adr/ADR-013.md
Mortezakoohjani e41ecfad4c Sync platform docs, infra, and module services with Accounting integration.
Include Loyalty/Communication/Sports Center backends and registry updates alongside production nginx and compose wiring.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-07-25 22:35:23 +03:30

2.2 KiB

ADR-013: Permanent AI Development Framework

Field Value
Status Accepted
Date 2026-07-24
Deciders Platform Architecture
Supersedes
Superseded by

Context

TorbatYar is implemented across many phases and services. AI-assisted implementation without a permanent process framework risks recreating completed work, violating module boundaries, skipping documentation gates, and drifting from ADRs.

Product AI (ai_assistant) is a business capability. Implementation process for any module must be governed separately.

Decision

  1. Adopt a permanent AI Development Framework under docs/ai-framework/.
  2. Every future implementation phase must follow the framework reading order, development loop, prompt rules, Cursor guidelines, templates, manifests, and quality gates.
  3. Phase prompts describe only the current phase; permanent rules live in the framework.
  4. No phase is complete until quality gates pass and phase-handover deliverables are filled.
  5. Machine-readable registration uses phase-manifest.yaml and service-manifest.yaml.
  6. This ADR does not implement product AI features.

Consequences

Positive

  • Consistent AI/human implementation lifecycle
  • Clear separation between process framework and product AI
  • Enforceable completion gates and handover continuity

Negative

  • Additional documentation discipline required before coding

Neutral

  • Existing architecture ADRs remain authoritative for technical decisions

Alternatives Considered

  1. Per-phase duplicated rules in every prompt — rejected (drift, inconsistency)
  2. Relying only on agent memory — rejected (not auditable)
  3. Treating product AI architecture as the implementation process — rejected (different concern)