@agent-native/core@0.145.8

Patch Changes

• c497c85: Stop running schema migrations on the serverless request path, and name every
  database connection.

  runMigrations now returns early in a production serverless request runtime.
  The guard lives in the shared runner rather than at each call site: the
  analytics template guarded its own runner, but org, context-xray, and
  observational-memory kept calling runMigrations unguarded, so the
  cold-start probe storm survived the fix meant to end it. Measured in
  production: the schema snapshot alone costs 5.5-8.6s on a 180-table database,
  with 4-6 copies running concurrently under load — and when it times out,
  ddl-guard falls back to a per-object probe across ~390 call sites, so
  starvation multiplies its own query count.

  A scheduled or background runtime claims migration duty through
  withMigrationRuntime(), and runInServerlessRequest: true remains the
  explicit opt-in for a caller that cannot defer. The database client also
  rejects unguarded schema DDL from production functions, so a new ensureTable
  path fails loudly instead of quietly reintroducing the incident.

  Better Auth table creation now lives in the framework's release migration
  entrypoint, and Analytics' production deploy runs its framework and template
  migrations once during the release build. No production request needs an
  environment variable to skip schema work.

  Postgres pools now set application_name. Every backend previously reported
  pgbouncer, which made a 58 MB SELECT id, config FROM dashboards running
  20-wide against production impossible to attribute — it appears nowhere in the
  repo or any built bundle, and pg_stat_statements is not installed.

• c497c85: Use layout-matching skeleton placeholders for shared settings loading states.