genesis-stack 1.4.6 → 1.4.8

package/README.md

@@ -33,7 +33,8 @@ executes catalog code. Framework pieces such as JSKIT and Vue link to their
 authoritative documentation in ordinary Stack guidance. Each piece's
 "Sanity checks with <stack>" guidance says how to apply Genesis's
 stack-neutral sanity checks with that stack's own tools, such as its strict
-loading mode, query-count assertion and parallel test runner; it names a
+loading mode, query-count assertion, query-side filters and parallel test
+runner; it names a
 framework default only where official documentation confirms it.
 
 Genesis still supports optional Agent Skills generically. A Stack piece from

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "genesis-stack",
-  "version": "1.4.6",
+  "version": "1.4.8",
   "description": "The optional curated technology Stack catalog for Genesis projects.",
   "type": "module",
   "repository": {

package/stacks/pieces/django.md

@@ -111,6 +111,9 @@ https://docs.djangoproject.com/
     rows.
   - `assertNumQueries` asserts a given number. Use it to compare a few rows
     with a realistic number, never as a fixed ceiling.
+  - Narrow on the queryset (`filter()`, `exclude()`, `exists()`, `count()`, a
+    slice taken before it is evaluated), never in Python: a comprehension or
+    loop over fetched rows, `len()` or `if queryset:` loads every row.
   - Django runs tests one at a time unless asked: run
     `manage.py test --parallel auto`, which clones the test database once per
     process, so the database user needs permission to create databases. Add

package/stacks/pieces/jskit-mysql.md

@@ -82,3 +82,11 @@ JSKIT MySQL runtime, live-schema CRUD generation, and persistence conventions.
   error translation at one established persistence boundary.
 - Never compete with or rewrite a generator-owned baseline migration. Preserve
   retained migration history and data semantics.
+
+## Data pass
+
+- Apply the MySQL part of JSKIT's data pass page at
+  `https://mobily-enterprises.github.io/jskit-ai/guide/framework/data-pass#mysql`:
+  one schema change per migration because MySQL DDL commits as it goes, online
+  index changes that fail rather than lock (`ALGORITHM=INPLACE, LOCK=NONE`),
+  and `EXPLAIN ANALYZE` on the test database at realistic size.

package/stacks/pieces/jskit-postgresql.md

@@ -78,3 +78,11 @@ JSKIT PostgreSQL runtime, live-schema CRUD generation, and persistence conventio
   error translation at one established persistence boundary.
 - Never compete with or rewrite a generator-owned baseline migration. Preserve
   retained migration history and data semantics.
+
+## Data pass
+
+- Apply the PostgreSQL part of JSKIT's data pass page at
+  `https://mobily-enterprises.github.io/jskit-ai/guide/framework/data-pass#postgresql`:
+  concurrent index builds in a migration without a transaction, constraints
+  added `NOT VALID` and validated afterwards on large tables, and
+  `EXPLAIN (ANALYZE, BUFFERS)` on the test database at realistic size.

package/stacks/pieces/jskit.md

@@ -93,7 +93,11 @@ https://mobily-enterprises.github.io/jskit-ai/
   describes; list, batch and write paths compare their query counts at
   realistic sizes, as
   `https://mobily-enterprises.github.io/jskit-ai/guide/framework/typed-features#query-counts-at-realistic-sizes`
-  describes.
+  describes. A read narrows in its query, with the filters the resource's
+  `searchSchema` maps (a field, an id list, `q`), never with `filter`, `find`
+  or `slice` on rows a query returned; a list's Search box and a lookup field's
+  autocomplete send `q` to the server, which narrows only where that resource
+  maps `searchSchema.q`.
 
 - When application integrations are requested or `integrations.json` exists,
   treat that file as the application-owned configuration used by both editor
@@ -322,6 +326,42 @@ https://mobily-enterprises.github.io/jskit-ai/
 - The checks cannot see layout, spacing or contrast; look at every screenshot
   for those.
 
+## Identity pass
+
+- Read JSKIT's visual identity page at
+  `https://mobily-enterprises.github.io/jskit-ai/guide/framework/visual-identity`
+  and its Material 3 rules at
+  `https://mobily-enterprises.github.io/jskit-ai/guide/framework/material-3`
+  completely. The identity lives where that page puts it: the seed and a
+  named colour per content kind in `src/theme/index.js`, every content colour
+  in `src/theme/palettes.js`, the cover and empty-state art in
+  `src/components/art/`, the logo with `favicon.svg`, the startup shell in
+  `index.html`, and sign-in in the auth surface's `AppBrand.vue`.
+- Run `npx jskit check m3` until it reports nothing and `npx jskit check`
+  until it passes, then take the screenshots as JSKIT's UI pass method shows.
+
+## Quality pass
+
+- Read JSKIT's quality pass page at
+  `https://mobily-enterprises.github.io/jskit-ai/guide/framework/quality-pass`
+  completely: where loading, empty, error and refused states live in a JSKIT
+  app, how each refusal carries its own code to `useJskitText()`'s
+  `serverMessage()`, and how Playwright forces each state.
+- Run `npm run verify` and the focused Vitest and Playwright tests of the
+  affected screens.
+
+## Data pass
+
+- Read JSKIT's data pass page at
+  `https://mobily-enterprises.github.io/jskit-ai/guide/framework/data-pass`
+  and its data access guide at
+  `https://mobily-enterprises.github.io/jskit-ai/guide/framework/data-access`
+  completely: the table resource is each table's one owner, Knex migrations
+  carry constraints and indexes, and `expectFlatQueries()` compares a path's
+  queries at a few rows and at a realistic number on `TEST_DB_NAME`.
+- Leave `npm run lint` (which reports narrowing in code) and `npx jskit check`
+  without findings.
+
 ## Verification
 
 - Verify `application`: `npm` `run` `verify`

package/stacks/pieces/laravel.md

@@ -104,6 +104,10 @@ conventions. Authoritative framework documentation: https://laravel.com/docs/
   - `get()` and `all()` return every row. `paginate()` returns one page, with
     `total` and `next_page_url` in its JSON, and never stands for every row.
     Batches iterate with `chunkById()` or `lazyById()`.
+  - Narrow in the query builder (`where`, local scopes, `whereHas`,
+    `exists()`, `count()`), never with a Collection's `filter()`, `where()` or
+    `first()` on fetched rows: `$post->comments()->where(...)` is a query,
+    while `$post->comments->where(...)` loads every comment and filters in PHP.
   - Laravel runs tests one at a time in one process. Add `brianium/paratest`
     as a development dependency and run `php artisan test --parallel`: each
     process gets its own `<database>_test_<token>` database, which Laravel

package/stacks/pieces/nodejs.md

@@ -73,6 +73,9 @@ Node.js package, command, module, dependency, and test conventions.
     `maxFailures` defaults to 0, no cap: set a low cap only on a diagnosis run.
   - Count a list's queries at a few rows and at a realistic number through the
     database client's query event or log.
+  - Narrow in the database client's query (its `where`, `limit` and count),
+    never with an array's `filter`, `find`, `some` or `slice` on rows already
+    read.
 
 ## Adoption
 

package/stacks/pieces/rails.md

@@ -132,6 +132,9 @@ Authoritative framework documentation: https://guides.rubyonrails.org/
     with a realistic number, never as a fixed ceiling.
   - `Model.all` has no `LIMIT`; batches iterate with `find_each`, 1,000 rows at
     a time.
+  - Narrow on the relation (`where`, scopes, `limit`, `exists?`, `count`),
+    never on loaded records: `length` and the block forms of `select`, `find`
+    and `any?` load every record and work in Ruby.
   - Keep the generated `parallelize(workers: :number_of_processors)` in
     `test/test_helper.rb`. Rails parallelizes from 50 tests up and creates a
     database for each worker, so the database user needs permission to create

package/stacks/pieces/spring-boot.md

@@ -102,7 +102,9 @@ https://docs.spring.io/spring-boot/
   and the persistence layer's own tools, as their official documentation
   describes them: count a list's statements at a few rows and at a realistic
   number, run test classes in parallel with a database each, and prepare test
-  databases without replaying every migration per test class.
+  databases without replaying every migration per test class. Narrow in the
+  repository query (a derived `findBy`, `existsBy` or `countBy` method,
+  `@Query`, a `Specification`), never with a stream `filter` over `findAll()`.
 
 ## Adoption
 

package/stacks/pieces/vue.md

@@ -45,6 +45,10 @@ conventions. Authoritative framework documentation: https://vuejs.org/
     production build has no such check, so repeat the changes there and look
     for a frozen tab or blank region. Count the DOM nodes a long list renders
     with realistic data.
+  - A picker, lookup or autocomplete over stored records sends the typed search
+    to the server and shows what it returns, never a loaded list filtered in a
+    `computed`. Vuetify's `v-autocomplete` filters only the items it holds:
+    give it `no-filter` and load its items on `update:search`.
 
 ### Reactive state hydration