Claude Code — LumiBase Agent Setup

Read docs/en/agent-setup/prompt.md and follow all setup instructions.

Fetch https://raw.githubusercontent.com/khuepm/lumibase/main/docs/en/agent-setup/prompt.md

cd /path/to/lumibase
claude

Read docs/en/README.md to understand the LumiBase monorepo structure.
Then read docs/en/data-model.md for the database schema.
Then read docs/en/ai-skills.md for the skill system conventions.

## LumiBase development rules

- Use NanoID or UUIDv7 for all IDs — never auto-increment serial
- Every domain table must have `site_id` for multi-tenancy
- Business logic goes through `@lumibase/runtime` abstractions (not direct CF bindings)
- Prefer 1-roundtrip aggregated API responses
- Dangerous AI actions (schema:write, delete*) must create ai_approvals rows
- Cache invalidation must use tag-based approach (not key enumeration)
- Follow the stack: Hono.js + Drizzle ORM + PostgreSQL + Cloudflare Workers / Docker

{
  "mcpServers": {
    "lumibase": {
      "url": "http://localhost:1989/mcp",
      "env": {
        "LUMIBASE_TOKEN": "<your-access-token>",
        "LUMIBASE_SITE_ID": "<your-site-id>"
      }
    }
  }
}

Read docs/en/features/flows-automation.md, then add a new "send-slack" operation type
to the operations engine. Follow the existing pattern for operation handlers.

Read docs/en/features/ai-copilot.md. Add a new AI skill called "publishCollection"
that sets a collection's status to "published". Mark it as safe (no HITL needed).

Read docs/en/data-model.md. Create a Drizzle migration to add a `tags` text[] column
to the items table, following the site_id multi-tenancy pattern.