Centralized Prompt Registry

You are a platform engineer. Build a centralized prompt registry for AI coding agents and LLM-powered features. Tech Stack: [INSERT - Next.js + FastAPI/NestJS + PostgreSQL] LLM: [INSERT - OpenAI / Anthropic / local models used for validation] Template Engine: [INSERT - Jinja2 / Handlebars / LangChain templates] Consumers: [INSERT - IDE plugin, CI bots, agent runtime] Deliver the following modules in order: 1. Architecture: versioned registry service plus renderer SDK; Git-backed storage so prompt changes ship through pull requests and review. 2. Data Model: prompt entity (key, version, template, variable schema, owner, tags, target model), immutable releases and aliases (prod/staging). 3. APIs: render endpoint with variable validation, batch render, release/promote/rollback, change webhooks; client SDKs with local caching. 4. Validation: variable schema validation, token-count checks, automated test-case runs and LLM-as-judge scoring on every prompt PR. 5. Governance: RBAC, approval flow, changelog, audit log, A/B split between versions, secret-variable injection. 6. Testing & Evals: golden output tests, rendering fuzz tests, drift detection across versions, performance tests for the render path.