A Command is a stably named function carrying "worldant::command" or the equivalent defineCommand() wrapper. Exporting the authored binding is optional.

import { session } from "worldant/client"

async function open(input) {
  "worldant::command"
  const reply = await session.request("pgpaw.sql", {
    sql: "select * from todo_items where done = false order by id",
  })
  return reply.rows
}

Commands do not receive a context argument. They import supported capabilities from worldant, and reach application data through the activation-bound session from worldant/client.

session.request("pgpaw.sql", { sql, params }) executes exactly one parameterized statement on the PgPaw data node and answers { command, rows, rowsAffected }. Each call commits independently: if the Command later throws, earlier commits remain committed. There is no handler-wide or multi-statement transaction surface — group effects that must commit together into one SQL statement (a CTE), or make the mutations idempotent.

Parameters bind as JSON against described types: cast non-JSON types via text (($1::text)::uuid), unnest JS arrays with jsonb_array_elements_text($1), and check reply.rows.length when zero rows is a normal outcome. The database page covers these gotchas with examples.