feat(tui)!: upgrade ink to 7.1.1

.changeset/connection-timeouts.md

@@ -0,0 +1,9 @@
+---
+"@noormdev/sdk": minor
+---
+
+## Connection
+
+### Added
+
+* `feat(db):` Connections now carry a connect timeout, so an unreachable host fails instead of hanging forever. Defaults to 15s; set `connection.connectTimeoutMs` per config to raise it for a database that resumes from an auto-paused state.

.changeset/tui-ink7-and-row-inspection.md

@@ -0,0 +1,24 @@
+---
+"@noormdev/cli": minor
+---
+
+## TUI
+
+### Added
+
+* `feat(tui):` The interactive UI now draws in the alternate screen at full terminal height, and restores the terminal on exit instead of leaving itself in scrollback.
+* `feat(tui):` Forms are a two-column layout with browse and edit modes. Arrow keys move between fields on every field type including selects, `Enter` opens a field and commits it, `Esc` reverts it, and submit is a `[ Save ]` row you navigate to.
+* `feat(tui):` Lists size themselves to the terminal instead of a fixed row count, and keep their cursor when you leave a screen and come back.
+* `feat(explore):` `r` on a table shows its first and last rows, ordered by primary key. `Enter` opens a row as YAML or JSON, `f` switches format, and the arrow keys walk between rows.
+* `feat(explore):` The detail screen scrolls, and `v` shows a value the column grid had to truncate.
+* `feat(sql):` Wide result grids drop whole columns behind a `… N more columns` marker rather than squeezing every column past legibility. `Enter` opens a row in full.
+* `feat(tui):` Mouse support: click to move the cursor, double-click to activate, wheel to scroll. Set `ui.mouse: false` in `.noorm/settings.yml` to restore click-drag text selection.
+* `feat(db):` `Esc` cancels a connection test or query that is hanging. On PostgreSQL and MySQL the server is asked to stop the query; on SQL Server and SQLite the client stops waiting and says so.
+
+### Fixed
+
+* `fix(sql):` Backspace in the results filter did nothing, because the key reports as `delete` on the previous Ink release.
+* `fix(tui):` Screens sized from the terminal never recomputed on resize.
+* `fix(explore):` Column, index and parameter lists re-flowed per row, so the type column landed on a different offset on nearly every row.
+* `fix(explore):` "Total Objects" counted categories the screen does not list, so it exceeded the rows a reader could see.
+* `fix(tui):` The help screen and log viewer drew past the bottom of the window, putting their first lines out of reach.

.claude/rules/documentation.md

@@ -1,45 +1,33 @@
 ---
-paths: docs/**/*.md
+paths:
+  - "docs/**/*.md"
 ---
 
-# Documentation Rules
+# Documentation rules
 
 
-## Three-Pillar Structure
+## Voice is assigned per file, not globally
 
-Build each section with Memory, Reasoning, and Example woven together. The pillars are invisible scaffolding.
+The `## Documentation surfaces` table in the root `CLAUDE.md` assigns every doc path a voice: `atomic-writing` or `terse-technical`. That table is the authority, and `/documentation` reads it to route authoring. Look the file up there before writing, and follow the `atomic-writing` skill when it says so.
 
-| Pillar | Purpose | Expression |
-|--------|---------|------------|
-| Memory | The what | Concepts, definitions, data structures |
-| Reasoning | The why | Motivation, trade-offs, design decisions |
-| Example | The how | Code samples, workflows, practical demonstrations |
+Do not impose one house tone across `docs/`. A CLI reference and a getting-started guide are labeled differently on purpose.
 
 
-## Style
-
-Start with the problem the reader wants to solve. Explain through analogy before technical detail.
-
-Use short sentences mixed with longer explanations. Show "why" before "how". Code follows explanation.
-
-Create visual hierarchy through headers, code blocks, and tables.
+## Claims
 
+Every claim has to be one you can point at.
 
-## Tone
+Do not invent a position to argue against. No "some people say", no "you may have been told", no opponent who does not hold the view. State the finding and let it stand without a foil.
 
-Conversational but precise. Guide the reader from familiar concepts to new ones.
+`docs/guide/changes/overview.md:15` currently breaks this ("Some argue that you *are* moving data..."). Fix it when you next touch that file, and do not copy the shape.
 
+Measurements carry their conditions: the tool, the scale, the run. Never combine numbers from different runs or builds into one comparison. If the conditions differ, re-measure or say which figure came from where.
 
-## Claims
+Mark inference as inference. An extrapolation from a mechanism reads as a measured result unless the text says otherwise.
 
-Every claim has to be one you can point at.
 
-Do not invent a position to argue against. No "some people say", no "you may have been told", no
-opponent who does not hold the view. State the finding and let it stand without a foil.
+## Structure
 
-Measurements carry their conditions — the tool, the scale, the run. Never combine numbers from
-different runs or builds into one comparison. If the conditions differ, re-measure or say which
-figure came from where.
+Lead with the problem the reader wants to solve, then the mechanism. Code follows the explanation that motivates it.
 
-Mark inference as inference. An extrapolation from a mechanism reads as a measured result unless
-the text says otherwise.
+Prefer a table, tree, or diagram wherever the content has a shape. Reserve prose for reasoning and motivation, which is the one thing a table cannot carry.

.claude/rules/testing.md

@@ -1,58 +1,95 @@
 ---
-paths: tests/**/*.{ts,tsx}
+paths:
+  - "tests/**/*.{ts,tsx}"
 ---
 
-# Testing Rules
+# Testing rules
 
 
-## Naming
+Runner is `bun:test`. `bun run test` is `bun test --serial`, and `bunfig.toml` pins `concurrency = 1`, `timeout = 30000`, and `preload = ["./tests/preload.ts"]`. All 315 test files import from `'bun:test'`; nothing here uses vitest or jest.
 
-Use `describe('module: feature', () => {})` format. Group by module, then by feature.
 
-```ts
-describe('runner: executeFile', () => {
+## Before running anything
 
-    it('should skip unchanged files', async () => {
+Integration tests need live databases on non-default ports (postgres `15432`, mysql `13306`, mssql `11433`):
 
-        // ...
+```bash
+docker compose -f docker-compose.test.yml up -d
+```
 
-    });
+`skipIfNoContainer(dialect)` (`tests/utils/db.ts:892`) **throws, it does not skip**, despite the name. 40 files call it in `beforeAll`. Without the containers up you get a wall of failures that read like real regressions. Check the containers before you believe a red suite.
 
-    it('should emit error event on failure', async () => {
 
-        // ...
+## Database safety
 
-    });
+The suite runs `TRUNCATE`, `DROP`, and teardown. `assertTestDatabase` (`tests/utils/db.ts:122`) refuses to connect unless the resolved database name is `:memory:` or contains `test` as a `_`/`-` delimited word, and `createTestConnection` calls it on every connection. It exists so a stray `.env` or leaked CI secret cannot aim a destructive suite at a real database.
 
-});
-```
+Never bypass it by building a Kysely instance by hand. Route test connections through `createTestConnection`.
 
 
-## Coverage
+## Naming
 
-Test all paths: success, error, edge cases. Verify observer events are emitted with correct data.
+`describe('module: feature', ...)` — 414 of 420 top-level describes follow this. Group by module, then by feature. Prefix `it` descriptions with "should".
 
 ```ts
-it('should emit error event on failure', async () => {
+describe('runner: executeFile', () => {
 
-    const events: any[] = [];
-    observer.on('file:after', (data) => events.push(data));
+    it('should skip unchanged files', async () => {
 
-    const [_, err] = await attempt(() => executeFile(badFile, configName));
+        // ...
 
-    expect(err).toBeInstanceOf(Error);
-    expect(events[0].status).toBe('failed');
+    });
 
 });
 ```
 
 
-## Error Handling
+## Errors
 
-Use `attempt` for operations that may fail. Assert both the error and any side effects.
+Use `attempt`/`attemptSync` and assert on the tuple. Assert the error *and* the side effects that should or should not have happened.
 
 ```ts
 const [result, err] = await attempt(() => executeFile(badPath));
+
 expect(err).toBeInstanceOf(InvalidFileError);
 expect(result).toBeUndefined();
 ```
+
+The no-`try/catch` rule in `typescript.md` covers `tests/**` too, and 17 blocks currently violate it. Do not add more.
+
+
+## Observer events
+
+Modules that emit events should have their emissions asserted, not just their return values. This applies to the ~20 files testing event-emitting modules (`core/change/executor`, `core/transfer/*`, `core/lifecycle/*`, and similar), not to the suite at large.
+
+```ts
+const events: ChangeEvent[] = [];
+observer.on('file:after', (data) => events.push(data));
+
+const [, err] = await attempt(() => executeFile(badFile, configName));
+
+expect(err).toBeInstanceOf(Error);
+expect(events[0].status).toBe('failed');
+```
+
+
+## Module mocking
+
+Bun's `mock.module` registry is process-global and never restores, so a mocking file poisons every file loaded after it. CI works around this by splitting the suite into five separate processes. The mechanics and the reasoning are in the root `CLAUDE.md`; read it before adding `mock.module` to a new file.
+
+
+## Helpers
+
+`tests/utils/db.ts` is the shared database harness:
+
+| Export | Purpose |
+|--------|---------|
+| `createTestConnection` | guarded connection; use instead of building Kysely directly |
+| `assertTestDatabase`, `NotATestDatabaseError` | the naming-convention safety guard |
+| `TEST_CONNECTIONS`, `makeTestConfig` | per-dialect connection details and a filled-in `Config` |
+| `deployTestSchema`, `seedTestData`, `resetTestData`, `teardownTestSchema` | fixture lifecycle |
+| `isContainerRunning`, `skipIfNoContainer` | container preflight |
+
+SQL fixtures live in `tests/fixtures/sql/<dialect>/`. `tests/integration/cli/setup.ts` spawns the built CLI for headless tests.
+
+`tests/global-setup.ts` and `tests/global-teardown.ts` are **dead code**, left over from vitest. Nothing wires them: `bunfig.toml` sets only `preload`, and there is no vitest config in the repo. Editing them changes nothing. Real preload work belongs in `tests/preload.ts`, which pins `NOORM_CHANNEL` and strips agent-harness markers so policy-gated tests behave the same locally and in CI.