mini/AI_ONBOARDING.md

🤖 AI Onboarding Protocol: Project Valhalla (God Mode)

Context: You are entering "God Mode", a high-performance, standalone Node.js system designed to bypass standard CMS limitations.

核心 Architecture (The "Truth")

• Repo: gatekeeper/mini.git (Standalone).
• Runtime: Node.js (Astro SSR adapter).
• Database: PostgreSQL (Directus Schema).
  • Note: We Do NOT run Directus. We map to its schema using raw SQL.
• Queue: Redis + BullMQ (BatchProcessor.ts).

⚡ Critical Systems (The "Shim")

File: src/lib/directus/client.ts

• Function: Intercepts standard SDK calls (readItems, createItem) and converts them to pg pool queries.
• Why: Allows "Headless" operation. If the API is down, we still run.

⚠️ "God Tier" Limits (The "Dangerous" Stuff)

File: docker-compose.yml

• Ulimit: 65536 (File Descriptors).
• Memory: 16GB (NODE_OPTIONS=--max-old-space-size=16384).
• Concurrency: 10,000 DB Connections.
• Warning: Do NOT lower these limits without checking the BatchProcessor throughput settings first.

🕹️ System Control (Standby Mode)

File: src/lib/system/SystemController.ts

• Feature: Standard "Push Button" to pause all heavy processing.
• Check: system.isActive() returns false if paused.
• Integration: BatchProcessor loops and waits if !isActive().

📜 Dictionary of Terms

• "Insane Mode": Running >50 concurrent threads.
• "Mechanic": Database Ops (src/lib/db/mechanic.ts).
• "Shim": The SQL conversion layer.
• "Proxy": The API route (/api/god/proxy) allowing the React Admin UI to talk to the Shim.