ash-data

Files (3)

• SKILL.md
• openai.yaml · agents/
• migrations-and-concurrency.md · references/

SKILL.md (first lines)

# Ash Data

State the persistence invariant that the change must preserve. Use installed
AshPostgres generation and migration behavior; inspect generated changes rather than
replaying archived migration history or assuming code generation only writes files.

[Migrations and concurrency](references/migrations-and-concurrency.md) covers
schema snapshots, constraints, atomic actions, transaction boundaries and SQL evidence.

Preserve full source records during imports, including unmapped columns. Identical
reimports must be harmless; conflicting identities or values need reconciliation.
A historical row count cannot prove present-day completeness. Rehearse on disposable
data before the separately authorized production operation.

A limit or once-only rule on an update is an atomic validation, never
`change filter(expr(...))`: Ash 3.33.11 drops that filter from single-record atomic
updates ([details](references/migrations-and-concurrency.md#four-distinct-concurrency-questions)).

Every site shares one production database: it serves through the pooler with a pool of
5 and listens on one direct connection per machine ([connections](references/migrations-and-concurrency.md#connections-to-the-shared-production-database)).

Do not disable atomic requirements reflexively. A race test needs independent
transactions, not two processes sharing one sandbox connection. Persistent retry or
locking guidance never authorizes admission state for user-signed Regent wallet sends.

From regents-ai/ash-template, MIT.

More from regents-ai/ash-template