@quickback-dev/cli 0.66.11 → 0.66.12

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (53) hide show
  1. package/dist/commands/deploy.d.ts.map +1 -1
  2. package/dist/commands/deploy.js +14 -0
  3. package/dist/commands/deploy.js.map +1 -1
  4. package/dist/index.js +1 -1
  5. package/dist/lib/api-client.d.ts +8 -0
  6. package/dist/lib/api-client.d.ts.map +1 -1
  7. package/dist/lib/api-client.js.map +1 -1
  8. package/dist/lib/deployment-target.d.ts +29 -0
  9. package/dist/lib/deployment-target.d.ts.map +1 -1
  10. package/dist/lib/deployment-target.js +26 -2
  11. package/dist/lib/deployment-target.js.map +1 -1
  12. package/dist/lib/deployment-target.test.js +37 -1
  13. package/dist/lib/deployment-target.test.js.map +1 -1
  14. package/dist/lib/migrations/drizzle.test.js +1 -1
  15. package/dist/lib/migrations/drizzle.test.js.map +1 -1
  16. package/dist/lib/migrations/plan.js +6 -0
  17. package/dist/lib/migrations/plan.js.map +1 -1
  18. package/dist/lib/migrations/postprocessors/better-auth-upgrade.d.ts +16 -5
  19. package/dist/lib/migrations/postprocessors/better-auth-upgrade.d.ts.map +1 -1
  20. package/dist/lib/migrations/postprocessors/better-auth-upgrade.js +32 -102
  21. package/dist/lib/migrations/postprocessors/better-auth-upgrade.js.map +1 -1
  22. package/dist/lib/migrations/postprocessors/index.d.ts.map +1 -1
  23. package/dist/lib/migrations/postprocessors/index.js +73 -1
  24. package/dist/lib/migrations/postprocessors/index.js.map +1 -1
  25. package/dist/lib/migrations/postprocessors/index.test.js +95 -0
  26. package/dist/lib/migrations/postprocessors/index.test.js.map +1 -1
  27. package/dist/lib/migrations/postprocessors/trigger-sql.d.ts +18 -0
  28. package/dist/lib/migrations/postprocessors/trigger-sql.d.ts.map +1 -1
  29. package/dist/lib/migrations/postprocessors/trigger-sql.js +63 -0
  30. package/dist/lib/migrations/postprocessors/trigger-sql.js.map +1 -1
  31. package/dist/lib/migrations/types.d.ts +1 -1
  32. package/dist/lib/migrations/types.d.ts.map +1 -1
  33. package/dist/lib/migrations/types.js +1 -0
  34. package/dist/lib/migrations/types.js.map +1 -1
  35. package/dist/skill/docs/define/actions/scoped-db.md +3 -0
  36. package/dist/skill/docs/define/schema.md +58 -6
  37. package/dist/skill/docs/define/triggers.md +143 -20
  38. package/dist/skill/docs/platform/auth/index.md +37 -16
  39. package/dist/skill/docs/platform/auth/plugins/subscriptions.md +25 -5
  40. package/dist/skill/docs/platform/database/neon.md +115 -18
  41. package/dist/skill/docs/platform/database/planetscale-postgres.md +3 -3
  42. package/dist/skill/docs/platform/storage/r2.md +75 -8
  43. package/dist/skill/docs/tooling/output.md +49 -5
  44. package/package.json +2 -1
  45. package/src/skill/docs/define/actions/scoped-db.md +3 -0
  46. package/src/skill/docs/define/schema.md +58 -6
  47. package/src/skill/docs/define/triggers.md +143 -20
  48. package/src/skill/docs/platform/auth/index.md +37 -16
  49. package/src/skill/docs/platform/auth/plugins/subscriptions.md +25 -5
  50. package/src/skill/docs/platform/database/neon.md +115 -18
  51. package/src/skill/docs/platform/database/planetscale-postgres.md +3 -3
  52. package/src/skill/docs/platform/storage/r2.md +75 -8
  53. package/src/skill/docs/tooling/output.md +49 -5
@@ -6,10 +6,20 @@ Each trigger lowers one of two ways, and the compiler reports which:
6
6
 
7
7
  | You write | Lowers to | Fires on |
8
8
  |---|---|---|
9
- | `sql:` | A real SQLite `CREATE TRIGGER` in your migrations | **Every** write path — REST, actions, queue handlers, cron, raw SQL. Atomic with the write. |
9
+ | `sql:` | A real database trigger in your migrations — a SQLite `CREATE TRIGGER` on D1, a `SECURITY INVOKER` PL/pgSQL function + trigger on Postgres | **Every** write path — REST, actions, queue handlers, cron, raw SQL. Atomic with the write. |
10
10
  | `handler:` | An application hook at the audit-wrapper write chokepoint | Writes through the wrapped `db` — REST CRUD and actions. **Not** queue handlers, cron, or raw SQL. |
11
11
 
12
- There is no auto-translation between the lanes: SQL bodies stay SQL, functions stay functions. Pick the lane by what the trigger must guarantee.
12
+ Both lanes exist on every database provider, with the same contract.
13
+ `handler:` is the default and the business-logic lane. `sql:` is the
14
+ **invariant** lane: data rules that must hold on every write path, atomically
15
+ with the write — soft-delete mapping, counters, denormalized fields. SQL
16
+ triggers carry data rules, never business flows and never role logic; the
17
+ compiler's own audit-stamping triggers are the precedent.
18
+
19
+ There is no auto-translation between the lanes — SQL bodies stay SQL,
20
+ functions stay functions — and none between SQL **dialects** either. A `sql:`
21
+ body is authored in the target database's own dialect and validated against
22
+ it at compile time. See [Dialects](#dialects-a-sql-body-is-never-translated).
13
23
 
14
24
  ## Basic Usage
15
25
 
@@ -66,30 +76,116 @@ Events: `beforeInsert`, `afterInsert`, `beforeUpdate`, `afterUpdate`, `beforeDel
66
76
 
67
77
  ## SQL triggers (`sql:`)
68
78
 
69
- The compiler wraps your statements in a `CREATE TRIGGER` named
70
- `qb_trg_<table>_<event>_<name>` and ships it in the journaled features
79
+ The compiler owns the object names and ships the DDL in a journaled
71
80
  migration — content-addressed, so an unchanged trigger set produces no new
72
81
  migration, and removing a trigger from config drops it from the database on
73
82
  the next migrate.
74
83
 
75
- - Write **bare, `;`-terminated statements** — the compiler owns the
76
- `BEGIN`/`END` frame. (It emits the frame uppercase; lowercase `begin`
77
- breaks remote D1's statement splitter even though local D1 accepts it.)
84
+ What is the same on every provider:
85
+
86
+ - You write **bare, `;`-terminated statements**. The compiler owns the
87
+ surrounding frame — you never author `CREATE TRIGGER` (or, on Postgres,
88
+ `CREATE FUNCTION`).
78
89
  - `OLD.column` / `NEW.column` references use **SQL column names**
79
90
  (`deleted_at`, not `deletedAt`) and are validated against the table at
80
91
  compile time.
92
+ - Optional `when:` is a SQL expression guard (`WHEN` clause).
93
+ - SQL triggers run inside the triggering statement, so they are atomic with
94
+ the write, and a rejection rolls the whole statement back.
95
+ - Names are `qb_trg_<table>_<event-code>_<name>` and are the compiler's, not
96
+ yours.
97
+
98
+ ### Dialects — a `sql:` body is never translated
99
+
100
+ A `sql:` body is raw SQL for **your** database. The compiler validates it in
101
+ the target's dialect and fails the compile when it sees the other one; it
102
+ never rewrites a body to fit. Declaring `sql:` on a provider with neither
103
+ lane (anything that is not `cloudflare-d1` or a Postgres provider) is also a
104
+ compile error.
105
+
106
+ > A SQLite `RAISE(ABORT, ...)` in a project targeting Postgres is rejected with:
107
+ >
108
+ > > `RAISE(ABORT, ...)` is SQLite syntax and has no PL/pgSQL equivalent — this
109
+ > > project targets PostgreSQL. Reject the write with `RAISE EXCEPTION
110
+ > > 'message';` instead. Trigger bodies are never translated between dialects.
111
+ >
112
+ > Porting a project between D1 and Postgres means porting its `sql:` bodies by
113
+ > hand. That is deliberate: a silent translation of a rule that must hold on
114
+ > every write path is the last thing you want to be wrong about.
115
+
116
+
117
+ #### SQLite (`cloudflare-d1`)
118
+
119
+ The body is inlined into `CREATE TRIGGER ... BEGIN <body> END`.
120
+
121
+ - The compiler emits the `BEGIN`/`END` frame **uppercase**; lowercase `begin`
122
+ breaks remote D1's statement splitter even though local D1 accepts it. An
123
+ authored frame is rejected rather than double-wrapped.
81
124
  - Bare `CASE ... END` expressions, including nested cases, are supported.
82
125
  For example: `SELECT CASE WHEN NEW.total < 0 THEN RAISE(ABORT, 'invalid total') END;`.
83
- That `END` closes the expression; authored `BEGIN ... END` trigger frames
84
- are rejected.
85
- - Optional `when:` is a SQL expression guard (`WHEN` clause).
86
- - To reject a write, `SELECT RAISE(ABORT, 'message');` — this is the only
87
- trigger form that blocks *every* write path, including raw SQL.
88
- - SQL triggers run inside the triggering statement — the only
89
- transactional option on D1, which has no `db.transaction()`.
126
+ That `END` closes the expression.
127
+ - To reject a write: `SELECT RAISE(ABORT, 'message');`
128
+ - SQL triggers are the only transactional option on D1, which has no
129
+ `db.transaction()`.
130
+
131
+ #### PostgreSQL (`neon`, `planetscale-postgres`)
90
132
 
91
- Currently supported on `cloudflare-d1` only; declaring `sql:` triggers on
92
- another database provider is a compile error.
133
+ Postgres has no inline trigger body, so each declaration becomes **two**
134
+ compiler-owned objects in the `app` schema: a `RETURNS trigger` function
135
+ carrying your statements, and the trigger that calls it.
136
+
137
+ ```typescript
138
+ triggers: {
139
+ beforeInsert: [
140
+ {
141
+ // Before-row mutation: assign to NEW. The compiler adds `RETURN NEW`.
142
+ name: 'defaultNotes',
143
+ sql: `NEW."notes" := COALESCE(NEW."notes", NEW."title");`,
144
+ },
145
+ ],
146
+ beforeUpdate: [
147
+ {
148
+ name: 'lockPaid',
149
+ when: `OLD."status" = 'paid' AND NEW."status" <> 'paid'`,
150
+ sql: `RAISE EXCEPTION 'paid invoices cannot be reopened';`,
151
+ },
152
+ ],
153
+ afterInsert: [
154
+ {
155
+ // Cross-table write. Unqualified names resolve in the `app` schema.
156
+ name: 'logCreate',
157
+ sql: `INSERT INTO invoice_events (id, invoice_id, organization_id, actor)
158
+ VALUES (NEW."id" || ':ins', NEW."id", NEW."organization_id", auth.user_id());`,
159
+ },
160
+ ],
161
+ }
162
+ ```
163
+
164
+ - Write **PL/pgSQL statements**, not a function. The compiler owns
165
+ `CREATE OR REPLACE FUNCTION`, the dollar quote, and the trailing
166
+ `RETURN NEW` (or `RETURN OLD` on a hard-delete trigger) — so a `before*`
167
+ trigger that mutates `NEW` proceeds, and you never return `NULL` by
168
+ accident and silently skip a write. Authoring `CREATE FUNCTION` or
169
+ `CREATE TRIGGER` is rejected.
170
+ - `IF … END IF`, `LOOP … END LOOP` and nested `BEGIN … END` blocks are
171
+ ordinary statements here and are accepted.
172
+ - To reject a write: `RAISE EXCEPTION 'message';` (SQLSTATE `P0001`). The
173
+ whole statement rolls back.
174
+ - Functions are **`SECURITY INVOKER`** and run with a pinned
175
+ `search_path = "app", public, pg_catalog`. That is the security contract:
176
+ a cross-table write from your trigger is still subject to RLS and grants,
177
+ so it cannot reach another tenant's rows — a cross-tenant write fails with
178
+ `42501` and rolls the statement back. Quickback never elevates a user
179
+ trigger to `SECURITY DEFINER`.
180
+ - The [caller claims](/platform/database/neon) are transaction-local, so
181
+ `auth.user_id()`, `auth.org_id()` and `current_setting('request.jwt.claim.sub', true)`
182
+ all read the acting principal from inside the body.
183
+ - Your triggers sort **before** the compiler's audit-stamping triggers
184
+ (`qb_trg_*` < `set_*_audit_on_*`, and Postgres fires BEFORE row triggers by
185
+ name), so a trigger cannot forge `createdBy` / `modifiedBy`.
186
+ - Object names are bounded to Postgres' 63-byte identifier limit with a hash
187
+ suffix, so two long trigger names that differ only past the cut stay
188
+ distinct instead of silently truncating onto each other.
93
189
 
94
190
  ### Soft-delete mapping
95
191
 
@@ -101,6 +197,26 @@ raw-SQL hard deletes are still covered. `*Update` events get the inverse
101
197
  guard, so a soft delete fires delete triggers exactly once and never
102
198
  double-fires as an update.
103
199
 
200
+ Identical on both dialects. On Postgres the two delete variants are two
201
+ functions, not one, because the transition trigger fires on `UPDATE` and
202
+ returns `NEW` while the hard-delete one returns `OLD`.
203
+
204
+ ### Upgrades and removal
205
+
206
+ The trigger migration is appended to the journal after the schema migration
207
+ (and, on Postgres, after the RLS document), and is content-addressed:
208
+
209
+ - An unchanged trigger set appends **nothing** on recompile.
210
+ - Editing or renaming a trigger appends one migration that drops the previous
211
+ objects and creates the current ones.
212
+ - Removing a trigger from config appends one drop-only migration; on Postgres
213
+ the generated **function** is dropped too, so no orphan is left behind.
214
+ - Nothing already applied is ever rewritten.
215
+
216
+ Drops never use `CASCADE` — an unexpected dependency fails loudly rather than
217
+ being swept — and Postgres trigger drops are guarded on `to_regclass`, so a
218
+ table the preceding schema migration removed does not break the cleanup.
219
+
104
220
  ## Handler triggers (`handler:`)
105
221
 
106
222
  Handlers run at the same write chokepoint that powers audit stamping and
@@ -161,15 +277,21 @@ Trigger cascades are bounded in every lane:
161
277
  - **Handler hooks**: the `db` a hook receives is re-wrapped at depth + 1.
162
278
  A cascade deeper than 5 levels (a hook's write firing hooks that write
163
279
  again, …) throws with the table/op that exceeded the cap.
164
- - **SQL triggers**: SQLite's `recursive_triggers` pragma defaults off — a
165
- trigger cannot re-enter itself, directly or through a cycle.
280
+ - **SQL triggers on D1**: SQLite's `recursive_triggers` pragma defaults off —
281
+ a trigger cannot re-enter itself, directly or through a cycle.
282
+ - **SQL triggers on Postgres**: there is **no** such guard. A trigger whose
283
+ write fires another trigger that writes back will recurse until Postgres
284
+ aborts on stack depth. Cascading `sql:` triggers on Postgres are yours to
285
+ keep acyclic — a `when:` guard on the state the trigger changes is the
286
+ usual way.
166
287
 
167
288
  ### Coverage — read the compile report
168
289
 
169
290
  Handler hooks do **not** fire for queue handlers, cron jobs, or raw SQL.
170
291
  The compiler emits a per-trigger lowering report and warns at compile time
171
292
  when a `before*` handler is used as a guard — if a rule must hold for every
172
- write path, use the `sql:` lane with `RAISE(ABORT)`.
293
+ write path, use the `sql:` lane and reject with `RAISE(ABORT, ...)` (D1) or
294
+ `RAISE EXCEPTION` (Postgres).
173
295
 
174
296
  > Cron jobs themselves are first-class — declare them with
175
297
  > [`defineSchedule`](/platform/schedules) and the compiler emits the Worker
@@ -180,10 +302,11 @@ write path, use the `sql:` lane with `RAISE(ABORT)`.
180
302
 
181
303
  | Need | Use |
182
304
  |---|---|
183
- | Invariant that must hold universally (raw SQL included) | `sql:` with `RAISE(ABORT)` |
305
+ | Invariant that must hold universally (raw SQL included) | `sql:` with `RAISE(ABORT)` (D1) / `RAISE EXCEPTION` (Postgres) |
184
306
  | Denormalized columns, counters, tombstones | `sql:` |
185
307
  | Validation/derivation with application logic | `before*` `handler:` |
186
308
  | Side-effects after a write (notify, write related rows) | `after*` `handler:` |
309
+ | Business flow, or anything that reads a role | `handler:` — `sql:` carries data rules only |
187
310
  | Atomic with the write (D1 has no transactions) | `sql:` only |
188
311
 
189
312
  Planned (not yet available — declaring them is a compile error rather than
@@ -70,25 +70,23 @@ role. Better Auth's own API is the gate for its tables.
70
70
  ## Compiler-managed auth versions and upgrades
71
71
 
72
72
  The compiler pins Better Auth, its schema CLI, and the generated companion
73
- packages to **1.7.2**. Recompile and install the emitted dependencies together;
73
+ packages to **1.7.3**. Recompile and install the emitted dependencies together;
74
74
  remove older Better Auth overrides from `build.dependencies`. An incompatible
75
75
  override is rejected because it would run a different auth version from the
76
76
  one that generated your schema.
77
77
 
78
78
  For an existing project, retain its `quickback/drizzle` migration history and
79
79
  use the updated CLI when recompiling. The CLI generates the upgrade locally;
80
- migration history stays on your machine. Apply the newly generated auth migration before
81
- starting the updated Worker. Existing migration SQL is preserved. The upgrade
82
- adds the current JWT/OAuth schema, including JWKS `alg`/`crv` and resource-bound
83
- grants, and backfills account identities without resetting credentials.
84
-
85
- Automatic account backfill covers password, SIWE, Google, Apple, Facebook,
86
- Line, GitHub, and compiler-managed Cloudflare accounts. Unknown/custom provider
87
- identities and duplicate identities must be resolved with an explicit migration;
88
- the generated migration fails before replacing existing account rows. Custom
89
- foreign keys into SQLite's account table also require an explicit migration.
90
- Rehearse the migration against a database copy, and stop old auth writers during
91
- the cutover so they cannot write the previous account shape.
80
+ migration history stays on your machine. Apply the newly generated auth
81
+ migration before starting the updated Worker. Existing migration SQL is
82
+ preserved.
83
+
84
+ ### Upgrading from 1.6
85
+
86
+ The upgrade adds the current JWT/OAuth schema, including JWKS `alg`/`crv` and
87
+ resource-bound grants. The `account` table is unchanged — accounts are still
88
+ keyed by `(providerId, accountId)`, so there is no identity backfill and no
89
+ credential reset.
92
90
 
93
91
  Existing OAuth grants without a recorded resource set are revoked during the
94
92
  upgrade; clients must authorize again. Client registrations and ordinary
@@ -99,7 +97,30 @@ grants still bind resources and reject widening to an unauthorized resource.
99
97
  Projects opting into per-client resource restrictions must link their clients
100
98
  to resources using the OAuth provider's resource administration APIs.
101
99
 
100
+ ### Projects compiled between 2026-09-05 and the 1.7.3 pin
101
+
102
+ Better Auth 1.7.0–1.7.2 briefly keyed accounts by an `issuer` column with a
103
+ unique `(issuer, accountId)` index. 1.7.3 restored the 1.6 `account` schema, so
104
+ only projects compiled during that window carry the extra column.
105
+
106
+ Recompiling emits one appended migration that drops the identity index and then
107
+ the `issuer` column — index first, because SQLite refuses to drop an indexed
108
+ column. It is generated from the schema diff, so a project that never had
109
+ `issuer` gets nothing, and prior migration SQL is untouched. No backfill is
110
+ needed in either direction: `issuer` was always derived from `providerId`, and
111
+ 1.7.3 never reads it.
112
+
113
+ Apply it before the updated Worker serves traffic. A leftover `NOT NULL`
114
+ `issuer` column is an outage on 1.7.3 — Better Auth validates its schema at
115
+ init and refuses a required column it never writes.
116
+
117
+ Generated projects keep that validation **on** at runtime
118
+ (`advanced.database.validateSchema: !!env` in `src/lib/auth.ts`). It is off in
119
+ exactly one case: when the schema CLI imports `auth.ts` there are no bindings,
120
+ so the adapter holds a stub with no tables to check and the validator would
121
+ report every table as missing.
122
+
102
123
  The bundled Account UI uses `signIn.social` and `linkSocial` for generic OAuth
103
- providers, including Cloudflare. Their callback is now
104
- `/auth/v1/callback/cloudflare`; update the registered redirect URI when upgrading
105
- an existing Cloudflare login integration.
124
+ providers, including Cloudflare. Their callback is
125
+ `/auth/v1/oauth2/callback/cloudflare`; update the registered redirect URI when
126
+ upgrading an existing Cloudflare login integration.
@@ -2,7 +2,7 @@
2
2
 
3
3
  The subscriptions plugin adds subscription tiers and Stripe billing to Better Auth. The compiler configures it from `quickback.config.ts`, emits it into your project as source, and wires its state-sync handlers onto the [inbound webhook surface](/platform/webhooks/inbound) — so Stripe events reach your database through one verified, deduplicated path.
4
4
 
5
- The subscriptions table lives in the auth database, alongside `user` and `session`: entitlements are an access-control concern, not domain data.
5
+ The subscriptions table lives alongside `user` and `session`: entitlements are an access-control concern, not domain data. See [Where the table lives](#where-the-table-lives) for what that means on each database provider.
6
6
 
7
7
  ## Enable it
8
8
 
@@ -61,7 +61,8 @@ These are compile errors, not warnings:
61
61
  - any tier after the first has no `priceId`
62
62
  - two tiers share a `priceId` (webhooks resolve a tier *by* price, so it would be ambiguous)
63
63
  - `webhooksBinding` is not set — see [Webhook wiring](#webhook-wiring)
64
- - the auth database is not Cloudflare D1 (the table ships in the SQLite dialect)
64
+ - the database provider has no subscriptions implementation — `cloudflare-d1`, `neon` and `planetscale-postgres` do
65
+ - the provider is Neon in the deprecated `connectionMode: 'websocket'`, whose database module emits no service-role handle
65
66
 
66
67
  ## Secrets
67
68
 
@@ -214,6 +215,25 @@ API keys and OAuth bearers stamp **nothing**, deliberately. A key is an integrat
214
215
 
215
216
  Org-vs-personal precedence applies here exactly as it does everywhere else — the active organization's subscription is what gets stamped. See below.
216
217
 
218
+ ## Where the table lives
219
+
220
+ One table, two dialects, selected at compile time. `src/plugins/subscriptions/schema.ts` re-exports whichever one your database provider uses, so drizzle-kit, the schema barrel and the runtime all resolve to a single definition.
221
+
222
+ | Provider | Table | Handle |
223
+ |----------|-------|--------|
224
+ | `cloudflare-d1` | `subscriptions` in the auth D1 database (`AUTH_DB`) | `drizzle(env.AUTH_DB)` |
225
+ | `neon`, `planetscale-postgres` | `auth.subscriptions`, in the same Postgres database and the same migration journal as everything else | the database module's internal service-role handle |
226
+
227
+ Column names, nullability and the JS values you read back are identical across the two. Timestamps are `integer` epoch-milliseconds on SQLite and `timestamptz` on Postgres; both hand your code a `Date`. `limits` is a JSON string on both.
228
+
229
+ ### On Postgres it is a SYSTEM table
230
+
231
+ `auth.subscriptions` carries the same shape as the webhook store and the rate-limit counters: RLS enabled and forced, an anonymous-denial policy, and the role bootstrap's service-role policy. **There is no caller scope policy**, so a request-scoped database handle reads zero rows from it — including the caller's own — and is denied every write with `42501`.
232
+
233
+ That is not a restriction on the API: every reader and writer of entitlement state already runs on the internal service lane. The Stripe queue consumer has no caller context at all, the admin endpoints run after a platform-admin check, and tier-role resolution runs inside the auth middleware. Authorization for the endpoints above is unchanged — session, admin role and organization membership, checked before the database is touched. What the missing scope policy buys is that a bug in an unrelated route cannot read or mint a paid entitlement through a caller handle.
234
+
235
+ Neon's deprecated `connectionMode: 'websocket'` is rejected for the same reason webhooks are: its database module emits no `createServiceDb`.
236
+
217
237
  ## Webhook wiring
218
238
 
219
239
  The plugin ships **no webhook route of its own**. Stripe events arrive on the standard inbound surface, which verifies the signature and deduplicates on `(provider, externalId)`, then dispatches through `WEBHOOKS_QUEUE`.
@@ -250,9 +270,9 @@ State sync runs on the queue, which is correct for entitlement propagation. Anyt
250
270
  ## What the compiler generates
251
271
 
252
272
  1. `src/plugins/subscriptions/lib/` — the plugin itself, emitted as source. It is **not** an npm dependency; the only package added to your `package.json` is `stripe`.
253
- 2. `src/plugins/subscriptions/schema.ts` — a re-export of the table in `lib/`, so drizzle-kit, the auth schema barrel and the runtime all resolve to one definition.
254
- 3. `src/plugins/subscriptions/index.ts` — the plugin with your tiers, auth database and org-membership lookup already bound.
255
- 4. A migration creating the `subscriptions` table in the auth database.
273
+ 2. `src/plugins/subscriptions/schema.ts` — a re-export of your dialect's table from `lib/`, so drizzle-kit, the schema barrel and the runtime all resolve to one definition. See [Where the table lives](#where-the-table-lives).
274
+ 3. `src/plugins/subscriptions/index.ts` — the plugin with your tiers, database handle and org-membership lookup already bound.
275
+ 4. A migration creating the subscriptions table — in the auth D1 database, or as `auth.subscriptions` in the single Postgres journal with service-role-only RLS.
256
276
  5. Registration of the Stripe handlers into the webhooks queue consumer.
257
277
  6. `STRIPE_SECRET_KEY` and `STRIPE_WEBHOOK_SECRET` on the generated `Env` type.
258
278
 
@@ -449,6 +449,20 @@ content-addressed `<idx>_quickback_rls_<hash>` migration, so one
449
449
  appends a new entry — recompiles with unchanged security append nothing. No
450
450
  manual SQL application step exists.
451
451
 
452
+ ### The default index set
453
+
454
+ Indexes are Drizzle's, not the security migration's: the compiler writes the
455
+ default set (tenant composite, FK columns, filtered columns, view-sortable
456
+ columns) into the schema source, so drizzle-kit generates and diffs them under
457
+ `<table>_<column>_idx` names. See
458
+ [The default index set](/define/schema#the-default-index-set) for the rules and
459
+ the `...q.indexes(t)` placeholder.
460
+
461
+ Older projects also carry `idx_<table>_<column>` copies the security migration
462
+ used to create. Those are dropped in the security migration's trailing
463
+ "retired indexes" section, which the journal runs after the schema migration
464
+ that created the replacements. Only `idx_*` names are dropped.
465
+
452
466
  Quickback also dependency-orders Drizzle's standalone constraint replacements.
453
467
  When a composite unique or primary key and its consuming foreign key are both
454
468
  replaced, the generated migration drops the dependent foreign key first, drops
@@ -530,15 +544,25 @@ owner's row. It is the floor under the application, not a second application.
530
544
  Quickback lowers your `firewall` config — and only your `firewall` config —
531
545
  into policies that read the transaction-local claims described above through
532
546
  `auth.user_id()`, `get_active_org_id()`, `get_active_team_id()`, `get_scope()`
533
- and the `get_principal_*()` helpers. Each feature table gets exactly four
534
- policies:
547
+ and the `get_principal_*()` helpers. Each feature table gets exactly three
548
+ policies, plus a fourth only where a delegated-principal lane is declared:
535
549
 
536
550
  | Policy | Kind | What it does |
537
551
  |--------|------|--------------|
538
- | `<table>_deny_anon` | RESTRICTIVE `FOR ALL` | Blocks any query carrying neither a stamped user id, nor a verified delegated/scope principal, nor the service marker |
552
+ | `<table>_deny_anon` | RESTRICTIVE `FOR ALL` | Blocks any query carrying neither a stamped user id, nor the service marker — nor, on a table with a delegated lane, a verified delegated/scope principal |
539
553
  | `<table>_service_role` | permissive `FOR ALL` | Admits Worker-internal contexts — see [Background jobs and cross-tenant actions](#background-jobs-and-cross-tenant-actions) |
540
- | `<table>_principal_fence` | RESTRICTIVE `FOR ALL` | Caps every permissive policy, so no policy can broaden a request that carries the wrong principal type |
541
- | `<table>_scope` | permissive `FOR ALL` | The firewall scope predicate. `USING` also requires `deleted_at IS NULL` on soft-deletable tables; `WITH CHECK` is the scope predicate alone, so a row can be soft-deleted but never resurrected |
554
+ | `<table>_scope` | permissive `FOR ALL` | The firewall scope predicate. On a soft-deletable table `USING` also requires `deleted_at IS NULL OR public.in_soft_delete()` — see [The soft-delete lane](#the-soft-delete-lane); `WITH CHECK` is the scope predicate alone, so a row can be soft-deleted but never resurrected |
555
+ | `<table>_principal_fence` | RESTRICTIVE `FOR ALL` | **Only on tables with a `databaseAccess` lane or `access.principals`.** Caps every permissive policy, so no policy can broaden a request that carries the wrong principal type |
556
+
557
+ The fence is per-lane, not per-table, because on a table no principal can reach
558
+ it would read `get_principal_type() IS NULL` — blocking every principal, which
559
+ is exactly what leaving them out of `deny_anon` already does. Two policies for
560
+ one outcome; the compiler emits neither.
561
+
562
+ Every helper call inside a policy predicate is emitted as `(SELECT …)`. The
563
+ claims are transaction-local and stamped before the statement, so PostgreSQL
564
+ hoists the call into an **InitPlan** and evaluates it once per statement instead
565
+ of once per scanned row.
542
566
 
543
567
  ```typescript
544
568
  firewall: [
@@ -552,13 +576,41 @@ ON "app"."documents" FOR ALL
552
576
  TO quickback
553
577
  USING (
554
578
  organization_id = public.get_active_org_id()
555
- AND deleted_at IS NULL
579
+ AND (deleted_at IS NULL OR public.in_soft_delete())
556
580
  )
557
581
  WITH CHECK (
558
582
  organization_id = public.get_active_org_id()
559
583
  );
560
584
  ```
561
585
 
586
+ ### The soft-delete lane
587
+
588
+ `deleted_at IS NULL` keeps soft-deleted rows out of every caller's reach. On its
589
+ own it also blocks the delete that put them there.
590
+
591
+ PostgreSQL applies a SELECT-applicable policy's `USING` expression to the **new**
592
+ row of an `UPDATE`, not only the existing one, whenever the statement reads the
593
+ relation — which any `UPDATE … WHERE` does. The new row of a soft delete has
594
+ `deleted_at` set, so a bare `deleted_at IS NULL` refuses the statement with
595
+ `42501`, including the generated `DELETE` route's own write. Moving the filter to
596
+ a `FOR SELECT` policy does not help, and neither does dropping `RETURNING`: the
597
+ check follows the filter wherever it goes.
598
+
599
+ So the filter carries a lane. `public.in_soft_delete()` reads the
600
+ transaction-local `quickback.soft_delete` marker, and the generated write
601
+ chokepoint sets it for **exactly** the statement whose payload stamps
602
+ `deletedAt` — the same condition that stamps the audit columns. It is a
603
+ one-shot token: the claims preamble carries it into that statement's
604
+ transaction and clears it, so the next statement of the same request runs
605
+ without it.
606
+
607
+ It is the same trust model as the service marker: Worker-set, never derived from
608
+ request input. And it is narrower — it exempts the soft-delete filter and
609
+ nothing else. Tenancy, ownership and scope stay ANDed beside it, so a statement
610
+ inside the window still cannot cross a tenant; `WITH CHECK` never carries the
611
+ lane, so it cannot land a row the firewall rejects; and the emitted statement
612
+ keeps `deleted_at IS NULL` in its own `WHERE`, so it cannot resurrect one.
613
+
562
614
  Exception tables (`firewall: [{ exception: true }]`) keep their permissive
563
615
  `<table>_all` policy, and SYSTEM tables — idempotency keys, the webhook store,
564
616
  Better Auth rate-limit counters, subscriptions, file metadata — stay
@@ -593,7 +645,8 @@ have to call still refuses them, and anything reaching the database outside a
593
645
  generated route holds a caller-scoped handle whose tenant is already fenced.
594
646
 
595
647
  The compiler enforces the shape at compile time. `QB-RLS-SCOPE-POLICY` requires
596
- all four policies, requires a tenant/owner/scope predicate in the scope policy's
648
+ the three policies (and the fence wherever a delegated lane exists — and only
649
+ there), requires a tenant/owner/scope predicate in the scope policy's
597
650
  `USING` clause, and **fails the build if any policy body calls `has_any_role`**.
598
651
  Opt out with `compiler.securityContracts.rls.requireScopePolicy: false` —
599
652
  `requireCrudPolicies` is the pre-016 name for the same key and is still honoured.
@@ -622,11 +675,14 @@ The retired names include the per-verb auth-table policies
622
675
  dropped, so an upgraded database converges on the narrower set rather than
623
676
  keeping a wider policy nobody emits any more.
624
677
 
625
- One leftover worth knowing: the redundant `idx_<table>_<column>` index the
626
- compiler used to infer alongside an index your schema already declared is no
627
- longer emitted, but an append-only security migration does not drop an index an
628
- older journal created. New databases get one index; existing ones keep the
629
- duplicate until you drop it yourself.
678
+ **Indexes are rebuilt once.** The compiler used to infer `idx_<table>_<column>`
679
+ indexes in the security migration, alongside the `<table>_<column>_idx` indexes
680
+ Drizzle generates from the schema. Drizzle now owns the whole default set, so
681
+ the schema migration creates the drizzle-managed index and the security
682
+ migration that follows it in the journal drops the compiler-owned copy — the two
683
+ overlap for the length of one `db:migrate` run and nothing is ever left
684
+ unindexed. Only `idx_*` names are dropped; a drizzle-managed index is never
685
+ touched. See [the default index set](#the-default-index-set).
630
686
 
631
687
  ### Delegated principal RLS
632
688
 
@@ -717,7 +773,7 @@ tenant predicate can never match), but it means a `databaseAccess` arm is never
717
773
  a place to write a restriction. A row that satisfies the arm is reachable even
718
774
  when the table firewall would not have matched it, and even when it is
719
775
  soft-deleted. Write the restriction in `firewall`; the restrictive
720
- `<table>_deny_anon` and `<table>_principal_fence` policies are the only arms
776
+ `<table>_deny_anon` and, where a lane is declared, `<table>_principal_fence` are the only arms
721
777
  that cap every lane. The fence also means a table admits a principal type only
722
778
  where its `access` (or its lane) names that type — a `ctx.principal.claims.*`
723
779
  firewall arm on a table whose access names no principal is unreachable.
@@ -937,6 +993,34 @@ precedent — so there is no separate `WEBHOOKS_DB` binding.
937
993
  (`requireOrgMember` / endpoint-ownership checks) before querying.
938
994
  - Webhooks require `connectionMode: 'http'`.
939
995
 
996
+ ## Managed file storage on Neon
997
+
998
+ `fileStorage: { managed: true }` is supported on `http` and `hyperdrive`. The
999
+ file metadata (`buckets`, `objects`) lives in a dedicated `files` Postgres
1000
+ schema of the **same** database — the `webhooks` precedent again — so there is
1001
+ no `FILES_DB` binding, no second drizzle config and no second migration target.
1002
+
1003
+ - The tables ride the project's single journal, reached through the
1004
+ `src/db/schema.ts` barrel.
1005
+ - Column shapes are 1:1 with the D1 version so no `/storage/v1/*` body changes:
1006
+ ISO-8601 date strings, JSON-encoded role and MIME lists, the same object keys
1007
+ and the same soft-delete column. The one divergence is numeric width —
1008
+ `size` and `file_size_limit` are `bigint`, because a Postgres `integer` would
1009
+ cap objects at 2 GiB.
1010
+ - The store is a SYSTEM table: RLS is enabled and forced with a single
1011
+ service-role policy, so only `createFilesDb()` (which wraps `createServiceDb`)
1012
+ can read or write it. There is no caller-scope policy and no role arm —
1013
+ public-versus-private file access is decided by the `/storage/v1/*` routes and
1014
+ the file-serving Worker before the handle is used, exactly as it was against
1015
+ the D1 binding.
1016
+ - The **file-serving Worker reads the same database over the same transport**:
1017
+ a `HYPERDRIVE` binding on Hyperdrive projects, or the same `DATABASE_URL`
1018
+ secret on HTTP projects — set on that Worker too. Never a mix of the two.
1019
+ - `connectionMode: 'websocket'` is rejected: it emits no `createServiceDb`.
1020
+
1021
+ See [File Storage](/platform/storage/r2) for the endpoints, bucket scopes and
1022
+ deployment steps.
1023
+
940
1024
  ## Encryption on Neon
941
1025
 
942
1026
  The encryption pillar — `.encrypted()` (envelope) and `.sealed()` /
@@ -1070,11 +1154,18 @@ Hyperdrive's connection pool.
1070
1154
  Anything not yet ported to Postgres is a compile-time error on Neon, never
1071
1155
  silently-broken output:
1072
1156
 
1073
- - **Managed file storage** (`fileStorage` with `managed: true`) is rejected — its
1074
- metadata store is a D1 database. Use presign-only R2 (the default, without
1075
- `managed`), which is provider-neutral and works on Neon today.
1076
- - **The Better Auth `subscriptions` plugin** is rejected until it gets
1077
- provider-aware db acquisition and a Postgres dialect schema.
1157
+ - **Managed file storage with `connectionMode: 'websocket'`** is rejected: the
1158
+ websocket db index emits no `createServiceDb`, and both `createFilesDb` and
1159
+ the file-serving Worker wrap that handle. Managed storage is supported on
1160
+ `http` and `hyperdrive` — the buckets/objects metadata lives in a `files`
1161
+ schema of this same database (one journal, no `FILES_DB` binding). See
1162
+ [File Storage](/platform/storage/r2).
1163
+ - **The Better Auth `subscriptions` plugin** is supported. Its table is created
1164
+ as `auth.subscriptions` in the same journal as everything else and is a SYSTEM
1165
+ table — service-role only, like the webhook store — because entitlement state
1166
+ is written by the Stripe queue consumer and the admin endpoints and read by
1167
+ tier-role resolution, all of which run on the internal service lane. Rejected
1168
+ on **WebSocket mode**, which emits no `createServiceDb`.
1078
1169
  - **Raw `sqliteTable` interop sources** on a Postgres target are rejected — write
1079
1170
  child tables with `pgTable` (see [below](#raw-drizzle-interop-tables-must-use-pg-core)).
1080
1171
  - **WebSocket mode** (deprecated) rejects webhooks, cross-tenant `unsafe`
@@ -1084,6 +1175,12 @@ silently-broken output:
1084
1175
 
1085
1176
  Mixed database providers (e.g. D1 auth + Neon features) remain unsupported.
1086
1177
 
1178
+ Table [`triggers`](/define/triggers) are supported in **both** lanes here. A
1179
+ `sql:` body is PL/pgSQL and lowers to a `SECURITY INVOKER` function plus its
1180
+ trigger in the `app` schema, so a cross-table write from a trigger stays
1181
+ subject to RLS and grants. Bodies are never translated between dialects — a
1182
+ SQLite `RAISE(ABORT, ...)` on this provider is a compile error.
1183
+
1087
1184
  ## Deploying
1088
1185
 
1089
1186
  ### Projects with no custom domain
@@ -181,9 +181,9 @@ never silently-broken output:
181
181
 
182
182
  | Feature | Status |
183
183
  |---------|--------|
184
- | Managed file storage (`managed: true`) | rejected — the file-metadata store is a D1 database |
185
- | Better Auth `subscriptions` plugin | rejected — its schema and db acquisition are D1-only |
186
- | Generated SQL triggers | not emitted on Postgres |
184
+ | Managed file storage (`managed: true`) | **supported** — metadata lives in a `files` schema of the same database; the file-serving Worker binds the same Hyperdrive config |
185
+ | Better Auth `subscriptions` plugin | supported — `auth.subscriptions` in the same journal, service-role only ([Subscriptions](/platform/auth/plugins/subscriptions#where-the-table-lives)) |
186
+ | Generated SQL triggers (`sql:`) | supported — PL/pgSQL bodies, `SECURITY INVOKER` (see [Triggers](/define/triggers)) |
187
187
  | `webhooksDatabaseId` / `webhooksDatabaseName` | rejected — D1-only keys; webhook tables live in the `webhooks` schema of the same database |
188
188
  | `connectionMode` of `'http'` or `'websocket'` | rejected — no such driver for this provider |
189
189