fix(engine): avoid Cloud migration ID collisions

CHANGELOG.md

@@ -31,7 +31,7 @@ Packages without a separate changelog are covered by the cross-package notes bel
 
 - Cloudflare D1 no longer rejects full 100-message history reads when attachment metadata is hydrated.
 - DM history query failures reach Relaycast Cloud's D1 overload handler and shared exception telemetry instead of becoming opaque route-level 500 responses.
-- Node enrollment no longer lets a workspace key alone rotate, rename, reshape, or reset an established node; existing rows require proof of their current node token while genuinely new node IDs remain enrollable. Keyed rotations retain a bounded, request-bound recovery record so a lost committed response returns the same replacement token on retry. Hosts must apply `0063_node_rotation_recovery.sql` before upgrading.
+- Node enrollment no longer lets a workspace key alone rotate, rename, reshape, or reset an established node; existing rows require proof of their current node token while genuinely new node IDs remain enrollable. Keyed rotations retain a bounded, request-bound recovery record so a lost committed response returns the same replacement token on retry. Hosts must apply `0066_node_rotation_recovery.sql` before upgrading.
 - Direct-message idempotency is now enforced atomically across direct and A2A routing, so concurrent retries reaching different server isolates cannot create duplicate rows or lose post-commit notification recovery.
 - Server telemetry events can no longer be emitted with an unlisted name, a missing required property, or an overridden `workspace_id`, any of which made hosted ingestion silently drop them.
 - Fix webhook listing responses so published SDKs can deserialize them, restoring webhook cleanup for integrations.

packages/engine/CHANGELOG.md

@@ -17,8 +17,8 @@ and this project follows [Semantic Versioning](https://semver.org/spec/v2.0.0.ht
 
 - DM, channel, thread, and delivery attachment hydration splits large message-ID lookups into D1-safe chunks, so full 100-message history pages stay within Cloudflare D1's bound-parameter limit.
 - `GET /v1/dm/:conversation_id/messages` lets query failures reach the global error handler for exception telemetry and host-specific error classification.
-- `POST /v1/nodes` requires an established node's current token in `X-Relaycast-Node-Token` before rotating or modifying the row; workspace-key-only enrollment remains valid for genuinely new nodes and returns `409 node_token_proof_required` without mutation for existing rows. A high-entropy `Idempotency-Key` binds the prior proof and exact request for 24 hours so a retry after a lost committed response recovers the same replacement token without another mutation. Hosts must apply `0063_node_rotation_recovery.sql` before upgrading.
-- `POST /v1/dm` admits one idempotency claim and durable notification state atomically across direct and A2A routing, preventing concurrent retries on different server isolates from creating duplicate messages or losing post-commit recovery. Hosts must apply `0062_direct_dm_idempotency.sql` before upgrading.
+- `POST /v1/nodes` requires an established node's current token in `X-Relaycast-Node-Token` before rotating or modifying the row; workspace-key-only enrollment remains valid for genuinely new nodes and returns `409 node_token_proof_required` without mutation for existing rows. A high-entropy `Idempotency-Key` binds the prior proof and exact request for 24 hours so a retry after a lost committed response recovers the same replacement token without another mutation. Hosts must apply `0066_node_rotation_recovery.sql` before upgrading.
+- `POST /v1/dm` admits one idempotency claim and durable notification state atomically across direct and A2A routing, preventing concurrent retries on different server isolates from creating duplicate messages or losing post-commit recovery. Hosts must apply `0065_direct_dm_idempotency.sql` before upgrading.
 - `emitServerEvent` only accepts `SERVER_TELEMETRY_EVENTS` names and requires each event's required properties at compile time; `workspace_id` always comes from the `workspaceId` argument.
 - `GET /v1/webhooks` returns `workspace_id`, `channel_id`, `channel_name`, and `created_by` for SDK compatibility while retaining `channel` for existing REST clients.
 

packages/engine/src/db/__tests__/compactMigrations.test.ts

@@ -54,7 +54,7 @@ function snapshot(handle: SqliteDbHandle) {
         const { status_updated_at: _statusUpdatedAt, ...existing } = row as Record<string, unknown>;
         return JSON.stringify(existing);
       }
-      // 0063 adds only bounded node-rotation recovery metadata. Exclude those
+      // 0066 adds only bounded node-rotation recovery metadata. Exclude those
       // nullable fields so this comparison remains byte-for-byte over every
       // pre-existing value while the explicit schema assertions below cover
       // the new columns.
@@ -118,7 +118,7 @@ function expectConstraintsPreserved(before: ReturnType<typeof constraints>, afte
       if (table.name === 'agents' && original) {
         return { ...table, sql: original.sql };
       }
-      // 0063 appends four nullable recovery fields. Remove only their exact
+      // 0066 appends four nullable recovery fields. Remove only their exact
       // additive DDL so changes to every pre-existing node constraint remain
       // visible to this preservation assertion.
       if (table.name === 'nodes') {

packages/engine/src/db/__tests__/taskMigration.test.ts

@@ -26,8 +26,8 @@ it('upgrades existing actions and results additively, preserving rows and every
     expect(runMigrations(handle).applied).toEqual([
       '0060_durable_task_invocations.sql',
       '0061_messages_workspace_length_id_index.sql',
-      '0062_direct_dm_idempotency.sql',
-      '0063_node_rotation_recovery.sql',
+      '0065_direct_dm_idempotency.sql',
+      '0066_node_rotation_recovery.sql',
     ]);
     for (const table of before) {
       const columns = db.pragma(`table_info(${table.name})`) as { name: string }[];