@qoretechnologies/reqraft 0.10.64 → 0.10.65-pr.137.g62fcecc

package/.tasks/QONSOLE_ASSIST_FEATURES.md

@@ -1,306 +0,0 @@
-# QonsoleSmartInput — assist response features
-
-Follow-up to the `0.10.0` UX batch (see
-[`SMART_EDITOR_UX_POLISH.md`](./SMART_EDITOR_UX_POLISH.md)). The Qorus
-LSP server's Qonsole assist response carries fields the client either
-silently discards or — in the case of `mode` — type-mistypes. This
-task triages each field, lands the generic ones in `useLspAutocomplete`,
-and lands the Qonsole-specific wizard / starter-suggestion flow in
-the `QonsoleSmartInput` wrapper.
-
-**Status:** done — committed in `189fe52`. Generic LSP wins (commitCharacters, sortText, warning) + Qonsole-specific (wizard launch via `qonsole.startWizard`, mode-type fix). **Awaiting browser verification.**
-**Scope:** seven items, three of them generic-LSP wins (commit chars,
-sortText, warning), four Qonsole-specific (relatedWizards, wizard
-command, natural-language mode, mode-type fix).
-**Estimated size:** ~200 lines + tests, ~1.5 days.
-
-## Goal
-
-A Qonsole consumer typing `/li ` should auto-accept `list` on space.
-Typing `/list ` should rank the resource list per server intent
-(group rank + match rank). Selecting a `Start wizard ...` item should
-launch the wizard flow, not insert the text. Typing free text (NL
-mode) should show the server's starter suggestions if available.
-Destructive verbs (`delete`, `drop`) should flag themselves in the
-dropdown so users see the warning before pressing Enter.
-
-None of these are blocking for the `0.10.0` release — the editor
-works without them. They're the difference between "Qonsole input
-typing works" and "Qonsole input feels like the IDE wrote it".
-
-## Research findings
-
-Source: `qorus/Classes/QonsoleAssistService.qc`,
-`qorus/Classes/QorusLspWebSocketHandler.qc`,
-`QONSOLE_LSP_REFERENCE.md`, plus a code search across our
-`useLspAutocomplete.ts`, `useQonsoleSession.ts`,
-`lspClient.types.ts`.
-
-### Generic LSP fields (apply to all languages, not Qonsole-only)
-
-These already exist on `ILspCompletionItem`:
-
-| Field | Server emits | Client uses | Severity |
-|---|---|---|---|
-| `commitCharacters: string[]` | yes (verbs/resources `[" "]`, non-bool flags `["="]`, bool flags `[" "]`) | **no** — typed but never read | Small but visible win |
-| `sortText: string` | yes (`<group_rank>_<match_rank>_<label>` shape, sorted server-side) | partial — preserved in items array, but grouping by kind in `useLspAutocomplete.ts:282` re-orders across groups | Medium — affects rank order |
-| `data: Record<string, unknown>` | yes (verb/resource/flag metadata; for wizard items: `{action: 'start-wizard', name, start_path, ...}`) | accessible via `item.raw.data` but no client codepath reads it | Small in isolation; required for wizard launch |
-
-### Qonsole-specific (only on this server / this language)
-
-| Field | Server emits | Client uses | Severity |
-|---|---|---|---|
-| `warning: string` | yes — only on mutating verbs (`create`, `delete`, `update`, …) | no — not in our type | Small but useful safety affordance |
-| `command: {command, arguments, title}` | yes — only on wizard completion items (`kind === 15`) | no | Required to make wizard items actionable |
-| `relatedWizards` (top-level on `qonsole/assist` response) | yes — when canonical command has both verb+resource AND user can access matching wizards | no — `IQonsoleAssistResult.related_wizards` exists in types but the field is never rendered | Small (wizards already surface as completion items via `makeWizardCompletionItems`) |
-| `mode: 'command' \| 'nl' \| 'empty'` | yes — server inferred from input | type bug: declared as `'command' \| 'natural-language'` in our types; server returns the literal `"nl"` | Tiny but correctness |
-
-### Server protocol details we need to know
-
-- **`workspace/executeCommand` is REJECTED for `qonsole.*`** — confirmed in `QorusLspWebSocketHandler.qc:2188-2207`. Wizard launch must go through REST: `POST {apiHost}{start_path}` from `command.arguments[0].start_path`, then drive the returned `QonsoleWizardSession` via `POST /api/latest/qonsole/wizards/sessions/{id}/step` and `/abort`. Documented in `QorusQonsoleRestClass.qc:7770-7950`.
-
-- **`completionItem/resolve` is NOT supported** by the server (`completionProvider.resolveProvider` is not in capabilities). `data` is fully populated up front — not for lazy resolution. So clients don't need to implement a resolve-roundtrip.
-
-- **Natural-language mode is detection-only.** The `nl` response has empty `completion.items`, empty `diagnostics`, optional `starter_suggestions` when the server's `isCannedWelcomeIntent` matches. Actual execution goes through `/qonsole/exec` SSE — out of scope for this task (it's the chat-runner, not the editor).
-
-## Design decisions
-
-### 1. `commitCharacters` — generic, lands in `useLspAutocomplete`
-
-VS Code semantics: when the focused item has `commitCharacters` and
-the user types one, accept the completion AND insert the typed
-character. The typed character is preserved in the document; the
-completion is what would have been typed had they pressed Tab/Enter.
-
-Special case `space` → bool flags: server emits `[" "]` for bool
-flags. After accepting `--verbose` on space, the server's
-`textEdit` already includes the space-after, so we'd double-space.
-**Decision:** on commit-char accept, check whether the inserted
-text from the completion already ends in the typed character — if
-so, suppress the verbatim re-insert. Confirmed by reading
-`QonsoleAssistService.qc:982`: flag insertText is `"--name="`
-(no trailing space) for `=`-committed flags, `"--name"` (no trailing
-char) for space-committed bool flags. So always re-insert is
-*almost* right, but the user types `--verbose ` and gets
-`--verbose ` (one space). For `=` flags, user types `--name=` and
-gets `--name=` (no double-equals because the completion's `newText`
-ends with `=`).
-
-Actually re-examining: the server's `textEdit.newText` is what the
-completion produces. For bool `--verbose`, server's `newText`
-is `--verbose` (without trailing). User types space → we accept →
-`--verbose` is inserted → we re-insert the typed space → final
-text is `--verbose ` (with one space). Correct.
-
-For `=` flag `--limit`, server's `newText` is `--limit=`. User
-types `=` → we accept → `--limit=` is inserted → we re-insert
-`=` → final is `--limit==`. **Wrong.** So we DO need the
-deduplication: if the just-inserted text ends with the commit
-character, swallow the second one.
-
-Implementation: peek at the completion's `textEdit?.newText ?? insertText ?? label`.
-If it ends with the typed key, suppress the re-insert.
-
-### 2. `sortText` — generic, lands in `useLspAutocomplete`
-
-Server already sorts; we re-order via kind-grouping which violates
-server intent.
-
-**Decision:** drop the kind-grouping for items whose `sortText` is
-present. Instead, group only when no `sortText` is supplied (legacy
-fallback). The server's `sortText` already encodes the group order
-(`group_rank` is the first segment). So we render as a flat list,
-ordered by `sortText`, with the server's headers implicit in the
-rank steps.
-
-This is a behaviour change for the dropdown when consuming Qonsole.
-For DPQL (which doesn't populate `sortText` consistently per server
-audit), groups remain. We branch on the presence of `sortText` on
-any item.
-
-Alternative we considered: keep groups but order group headers by
-the smallest `sortText` per group. Rejected — `sortText`'s first
-segment is `group_rank` exactly so a flat sort is equivalent and
-simpler.
-
-### 3. `warning` — generic field on `ILspCompletionItem`
-
-**Decision:** add `warning?: string` to `ILspCompletionItem`. Add
-`warning?: string` to `ICompletionDropdownItem`. Render as a small
-right-aligned warning chip + the row's `tooltip` payload extended
-with the warning. Visual: `<ReqoreTag intent='warning' icon='AlertLine' size='small'>` in the row's `rightAction` or as a second badge.
-
-Doesn't conflict with the kind-chip badge — Reqore allows `badge` to be an array.
-
-### 4. `data` — generic field on `ILspCompletionItem`
-
-**Decision:** add `data?: Record<string, unknown>` to
-`ILspCompletionItem`. No special rendering — it's a pass-through
-for inserters and external callers. The wizard-launch path (next
-item) reads `data.action === 'start-wizard'` to branch.
-
-### 5. Wizard launch via `command: 'qonsole.startWizard'` — Qonsole-specific
-
-Spec: the server attaches a `command: { command, arguments, title }`
-to completion items when `data.action === 'start-wizard'`. We need
-to NOT insert text but instead fire a callback the consumer wires
-to the wizard runner.
-
-**Decision:**
-
-- Generic addition to `ILspCompletionItem`: `command?: { command: string; title: string; arguments?: unknown[] }`.
-- Generic addition to `TCompletionInserter` context: we already pass
-  `replacementPath` etc. — add nothing; the inserter receives
-  `item.command` via `item.raw.command` and decides what to do.
-- Qonsole-specific addition: a new `onWizardStart?(args: unknown[])` prop
-  on `QonsoleSmartInput`. The default `completionInserter` for
-  Qonsole checks `item.raw.command?.command === 'qonsole.startWizard'`
-  and calls `onWizardStart(args)` instead of inserting text.
-- Reqraft provides NO wizard runner UI — that's qorus-ide-side
-  (a new `<QonsoleWizardSessionModal>` component). Our job is the
-  callback hand-off.
-
-The dropdown visual for wizard items uses a distinctive icon and a
-`Wizard` kind chip — extend the kindLabel map to map "snippet" kind
-to "Wizard" when `data.action === 'start-wizard'`.
-
-### 6. `relatedWizards` (top-level on assist response) — Qonsole-specific
-
-The server appends each wizard ALSO as a synthetic completion item
-(`makeWizardCompletionItems` at `QonsoleAssistService.qc:806`) — so
-the items list already covers the wizard surface. The top-level
-`related_wizards` array is the canonical source for an inline
-"Suggested setup" callout panel below the input — useful when a
-command is recognised but partially filled.
-
-**Decision (v1):** rely on the synthetic items in the dropdown. Do
-NOT build the inline callout panel. If we want it later, it's a
-new ReqoreCallout below the editor body when
-`session.lastAssist?.related_wizards?.length > 0`. Tracked as
-deferred.
-
-### 7. `mode` type fix
-
-Trivial — change `'natural-language'` to `'nl'` in
-`qonsoleSmartInput/types.ts:40`. Add `'empty'` as a third literal.
-
-Also fix `QONSOLE_LSP_REFERENCE.md:150` which has the same wrong
-literal documented.
-
-### Not landing in this task
-
-- Natural-language mode **starter suggestions** rendering — different
-  surface (chips below the input). Add later if Qonsole NL becomes a
-  Reqraft consumer's actual use case. The qorus-ide existing
-  QonsoleInput.tsx doesn't even use `qonsole/assist` for NL — goes
-  straight to `/qonsole/exec` SSE on submit.
-- Attachment-mode UX — assist response carries `attachments` only in
-  NL mode; out of scope.
-- The wizard runner modal itself — qorus-ide work.
-
-## Surface area
-
-| File | Change |
-|---|---|
-| `src/utils/lspClient.types.ts` | `ILspCompletionItem` gains `warning?: string`, `data?: Record<string, unknown>`, `command?: { command, title, arguments? }` |
-| `src/components/smartEditor/useLspAutocomplete.ts` | `ICompletionDropdownItem` gains `warning?`, surface `command` via `raw`; keyboard handler intercepts commit-char keys; sort by `sortText` (flat) when present; deduplication for commit-char re-insert |
-| `src/components/smartEditor/SmartEditor.tsx` | render the `warning` chip on rows that have it (right-side, intent="warning", small) |
-| `src/components/qonsoleSmartInput/types.ts` | fix `mode` literal (`'nl'`/`'empty'`); add `onWizardStart?(args: unknown[]) => void` prop |
-| `src/components/qonsoleSmartInput/QonsoleSmartInput.tsx` | inject a custom `completionInserter` that branches on `item.command?.command === 'qonsole.startWizard'` and calls `onWizardStart` |
-| `src/components/qonsoleSmartInput/qonsoleInserter.ts` | **new** — extracts the wizard-or-text branch into a named helper |
-| `__tests__/smartEditor/commitCharacters.test.ts` | **new** — unit tests for commit-char accept logic incl. the deduplication |
-| `src/components/qonsoleSmartInput/QonsoleSmartInput.stories.tsx` | enrich the mock to emit `warning` on destructive verbs and `command` on wizard items; add `WithCommitCharacters`, `WithWizardItems` stories |
-| `QONSOLE_LSP_REFERENCE.md` | fix the `mode` literal (`nl` not `natural-language`) |
-
-## Tasks
-
-### Generic field surfacing
-
-- [ ] `ILspCompletionItem` extended with `warning?`, `data?`, `command?`
-- [ ] `ICompletionDropdownItem` extended with `warning?`; populate it
-      in the mapping step of `performCompletionRequest`
-- [ ] SmartEditor renders the warning chip on the row's
-      `rightAction` when present
-
-### Commit characters
-
-- [ ] Extend the keydown handler in `useLspAutocomplete` to intercept
-      single-character keys that match the focused item's
-      `commitCharacters`; accept the completion via `selectItem`,
-      then re-insert the typed char UNLESS the just-inserted text
-      ended with it (server textEdit dedup)
-- [ ] Confirm interaction with `metadata.retrigger` — `:` commit on
-      a `$template:` item should re-trigger value completions; the
-      timeout-based re-trigger already handles this
-- [ ] Unit tests covering:
-  - bool flag (`--verbose`, space commit) → final text ends with one space
-  - `=` flag (`--limit`, `=` commit) → final text is `--limit=` (no double `=`)
-  - retrigger interaction → completion re-fires after commit
-
-### Sort
-
-- [ ] In `useLspAutocomplete`, sort `items` by `sortText` (lexically)
-      before grouping
-- [ ] When ANY item has `sortText`, skip kind-grouping — render a
-      flat list. When NO items have `sortText`, fall back to the
-      existing kind-grouping (preserves DPQL behaviour)
-
-### Qonsole-specific wizard flow
-
-- [ ] `qonsoleSmartInput/qonsoleInserter.ts` — new file with the
-      branch: wizard items call `onWizardStart`; everything else
-      delegates to `defaultCompletionInserter`
-- [ ] Add `onWizardStart?` prop to `IQonsoleSmartInputProps`
-- [ ] Hook the inserter via the `completionInserter` prop on
-      SmartEditor
-- [ ] In `useLspAutocomplete`, extend `ICompletionDropdownItem`'s
-      kindLabel resolver: when `kind === 15` (CIK_SNIPPET) AND
-      `data?.action === 'start-wizard'`, label the chip "Wizard"
-      and override the icon to `MagicLine`
-
-### Mode type fix
-
-- [ ] `qonsoleSmartInput/types.ts` — change `mode: 'command' | 'natural-language'` to `'command' | 'nl' | 'empty'`
-- [ ] Update `QONSOLE_LSP_REFERENCE.md` to match (and add `empty`)
-- [ ] No code change needed in `useQonsoleSession` — it passes the
-      response through unchanged
-
-### Stories + mock enrichment
-
-- [ ] `QonsoleSmartInput.stories.tsx` mock — for the BasicMock
-      completion response:
-  - Add `warning: 'Mutates system state'` to verbs `create`, `delete`,
-    `drop`, `update`
-  - Add `command: { command: 'qonsole.startWizard', title, arguments: [{action: 'start-wizard', name: 'create-connection', start_path: '/api/latest/qonsole/wizards/create-connection/start'}] }` to one synthetic wizard item per response
-  - Add `commitCharacters: ['=']` to flag completions like
-    `--limit`, `--search`, `--app`
-- [ ] New story `WithCommitCharacters` — initial value `/list services `,
-      play test types `--li`, then `=`, asserts the editor shows
-      `/list services --limit=` and the popover is open for values
-- [ ] New story `WithWizardItems` — initial value `create con`, mock
-      returns the wizard item, play test confirms selecting it fires
-      the consumer-provided `onWizardStart` callback (via `fn()` spy)
-
-### Documentation
-
-- [ ] README: add a short Qonsole-features section under the
-      QonsoleSmartInput entry mentioning commit chars, wizard launch,
-      `onWizardStart` prop
-- [ ] TSDoc on `onWizardStart` explains it's the consumer's
-      responsibility to drive the REST `start_path`
-
-### Checks
-
-- [ ] Typecheck, lint, jest, storybook play tests all pass
-- [ ] **STOP — user verifies in browser** before commit
-
-## Out of scope
-
-- Natural-language mode UX (starter cards, chat-style rendering) —
-  separate task if Qonsole NL becomes a Reqraft consumer use case
-- Wizard runner modal component — qorus-ide-side
-- `relatedWizards` inline callout panel — deferred; synthetic items
-  in the dropdown already cover the surface
-- `completionItem/resolve` plumbing — server doesn't advertise the
-  capability; no need
-- Attachment-mode handling

package/.tasks/RENDER_EXPRESSION_LSP.md

@@ -1,159 +0,0 @@
-# TASK — server "Explain" over the LSP (`dpql/renderExpression`)
-
-**Status:** code complete on branch `feature/render-expression-lsp`
-(uncommitted) — server rebuilt/restarted by the user 2026-06-10 and
-**verified live end-to-end** (probe + real-browser run of `LiveExplain`
-against :8012: preview AND Explain show `"test".startsWith("t", true)`
-served over `dpql/renderExpression`; the hq accidental run proved the
-`-32601` fallback path too). Awaiting user browser verification.
-**STOP — no commits until the user verifies.**
-
-Design: `design/RENDER_EXPRESSION_TRANSPORT.md` (locked 2026-06-10).
-Closes **Phase D** of `.tasks/EXPRESSION_BUILDER_REPORT.md`.
-Server ask history: `/tmp/render-expression-exposure.md`.
-
-## Surface area
-
-| Area | Path | Change |
-|---|---|---|
-| Server handler | `qorus/Classes/QorusLspWebSocketHandler.qc` | NEW `dpql/renderExpression` dispatch case + `handleDpqlRenderExpression()` (modeled on the creator's `handleIdeRenderExpression`, JSON-RPC like `handleDpqlToRichtext`) |
-| Server design doc | `qorus/design/language-server-api-metadata.md` | documented under "Custom Qorus LSP Methods" |
-| The seam | `src/components/form/expressions/useRenderExpression.ts` | server-first via shared module-level `LspClient`; client-side fallback kept; per-hook ~300ms debounce; `serverRendering` now live |
-| Mock LSP | `src/components/form/expressions/dpqlMockLsp.ts` | `dpql/renderExpression` case mirroring the server contract |
-| Stories | `Foundation.stories.tsx` (mock in `beforeEach`), `ExpressionField.stories.tsx` (NEW `LiveExplain` live story) | |
-| Storybook config | `.storybook/main.ts` | forward `REACT_APP_QORUS_INSTANCE`/`_TOKEN` shell overrides to the bundle (live-story flow was silently broken) |
-| Untouched consumers | `builder/renderTemplate.tsx`, `ExpressionField.tsx` | public hook contract preserved |
-
-## Phases
-
-### Phase 0 — shortcut check (DONE 2026-06-10)
-
-- [x] Probed live `wss://localhost:8012/lsp`: `dpql/serialize` richtext =
-      DPQL string in paragraph nodes; `dpql/toRichtext` takes `text` only;
-      `dpql/explain` / `dpql/renderExpression` → `-32601`.
-- [x] Probed live `wss://localhost:8012/creator` `render-expression`:
-      `"test".startsWith("t", true)` for the same AST — semantic rendering,
-      ≠ serialize output. **Shortcut disproven → server change needed.**
-- [x] Learned: creator accepts ONLY the `{is_expression, value}` wrapper
-      (bare AST → `EXPRESSION-ARG-ERROR`); `and` is not in
-      `QorusExpressionMap` (serializer accepts it, renderer doesn't).
-      Probe scripts: `/tmp/render-step0.mjs`, `/tmp/render-step0-creator.mjs`.
-
-### Phase 1 — server (code DONE; deploy PENDING)
-
-- [x] `handleDpqlRenderExpression()` in `QorusLspWebSocketHandler.qc`:
-      accepts bare AST or wrapper, optional `expmap`; returns
-      `{ rendered, richtext }`; stateless (no uri/session).
-- [x] `qorus/design/language-server-api-metadata.md` updated.
-- [x] **Committed + MR'd (2026-06-11):** qorus branch
-      `feature/render-expression-lsp`, commit `bccdb971c`, pushed with
-      `-o ci.skip` (push pipeline skipped; repo has no MR pipelines) —
-      https://git.qoretechnologies.com/qorus/qorus/-/merge_requests/2103,
-      assigned to dnichols.
-- [x] **Rebuilt + restarted by the user (2026-06-10).** Probe re-run green:
-      `dpql/renderExpression` returns
-      `{ rendered: "\"test\".startsWith(\"t\", true)", richtext: … }` for
-      both the bare AST and the wrapper, with or without `uri`; missing
-      expression → `-32602`; `and` (not in expmap) → clean `-32803`;
-      nested `&&` → `("test".startsWith("t", true)) && (5 > 3)`.
-      Note: qorus-core restarts invalidate session tokens — the probe
-      script now logs in itself (`admin`, local dev creds).
-
-### Phase 2 — reqraft seam swap (DONE)
-
-- [x] `useRenderExpression`: lazy module-level `LspClient` shared by all
-      consumers (rides `LspSharedConnection` — same socket as open DPQL
-      editors); JSON-RPC id association; per-hook leading+trailing ~300ms
-      debounce; bounded connect wait (1.5s) + 10s fail cooldown; `-32601`
-      remembered → permanent client-side fallback on old servers;
-      `serverRendering` flag now reflects the serving path live.
-- [x] Public API unchanged (`{ render, serverRendering }`,
-      `renderExpressionToText` still exported); consumers untouched.
-
-### Phase 3 — tests (DONE)
-
-- [x] `dpqlMockLsp.ts` extended with `dpql/renderExpression`.
-- [x] `Foundation.Default` now runs the mock (its `findByText` 1s timeout
-      is shorter than the offline fallback wait — would have gone flaky).
-- [x] NEW `ExpressionField.LiveExplain` story (`tags: ['!test']`,
-      `parameters: { live: true }`) for verification against :8012.
-- [x] **GATE:** eslint clean · `build:test:prod` clean · jest **268/268** ·
-      expressions play tests **17/17** · full storybook suite **200/201**
-      (only the pre-existing `FormEngine › OnValidityChange` flaky).
-
-### Phase 4 — verification (in progress)
-
-- [x] Server rebuilt + restarted, step-0 script green (2026-06-10).
-- [x] Agent end-to-end check: real chromium on the `LiveExplain` story at
-      :6011 (storybook pointed at :8012) — WS frames confirmed
-      `dpql/renderExpression` over the shared socket; preview AND Explain
-      show `"test".startsWith("t", true)`. The earlier hq-pointed run
-      doubled as the fallback proof (`-32601` → client-side rendering).
-      Found+fixed en route: `.storybook/main.ts` never forwarded
-      `REACT_APP_QORUS_INSTANCE`/`_TOKEN` from the shell, so the documented
-      live-story flow silently fell back to hq.
-- [x] **Styled Explain (2026-06-10, after user feedback):** the builder's
-      Explain now renders the server's `richtext` (template refs as chips,
-      via `richtextResponseToSlate` → `RichTextFormField`) — IDE parity;
-      hook gained additive `renderRich()` (`render` keeps its string
-      contract); fallback path now shows a muted "Approximate — server
-      rendering unavailable" tag so the serving path is never ambiguous.
-      Browser-verified on :6011: chips render, no badge, server text;
-      gates re-run green (jest 268/268, expressions play 17/17, eslint,
-      `build:test:prod`).
-- [x] **LSP-highlighted Explain + Text-seeding fix (2026-06-10, user
-      feedback round 2):**
-      (a) Both Explain surfaces (builder panel + ExpressionField Text-mode
-      Explanation) now render through a **read-only `DpqlEditor`** —
-      template-ref chips + LSP semantic-token colours; new
-      `showDiagnostics` prop on SmartEditor/DpqlEditor suppresses
-      underlines AND the message panel (the rendering is readable text,
-      not parseable DPQL — probe confirmed clean tokens but error diags).
-      Fallback keeps plain rich-text + the "Approximate" tag.
-      (b) Fixed ExpressionField Text-mode seeding: `enterTextMode` called
-      `dpqlRef.current.serialize` before the editor mounted (ref null →
-      silently empty editor); seeding now runs in an effect with a brief
-      retry until the editor's LSP session is ready; user typing aborts
-      a late seed. Browser-verified: 7 token colours + 6 chips in the
-      builder explain, editor seeds `"test" startsWith "t"`, re-seeds on
-      Visual→Text toggle. Full suite **201/201**, jest 268/268.
-- [x] **SERVER BUG found + root-caused (round-trip, case-sensitivity):**
-      `dpql/serialize` emits infix word operators (`"test" startsWith "t"`)
-      that `dpql/parse` rejects. Root cause in **qore** qlib
-      `DataProvider/DpqlParser.qc`: the comparison loop lowercases the
-      identifier (`string kw = tok.value.lwr();`, ~:964) before the infix
-      operator lookup `symbol_to_name{kw}` (~:1028), but the map is built
-      with symbols as-is (`symbol_to_name{"startsWith"}`, ~:136) — hash
-      lookup is case-sensitive, so camelCase symbols can never match.
-      PROOF: `"a" like "b"` (lowercase symbol) parses fine through the
-      same path (`exp=like`); `startsWith("test","t")` function form also
-      parses (`exp=starts-with`); only infix camelCase fails. The infix
-      path was added in qore commit `219938f44`. One-line fix in qore:
-      build the map with lowercased keys (`symbol_to_name{sym.lwr()} =
-      ename`) — matches DPQL's case-insensitive keywords (BETWEEN/IN).
-      Affects `startsWith`/`endsWith`/any camelCase-symbol comparison op.
-      Repo: QoreTechnologies/qore (NOT qorus) — needs qore qlib
-      reinstall; running copy is `~/omq/qlib/DataProvider/DpqlParser.qc`
-      (identical to source). Shows as a red diagnostic under the seeded
-      Text editor in `LiveExplain` until fixed.
-      **FILED + FIXED (2026-06-11):** issue
-      https://github.com/qoretechnologies/qore/issues/5348; fix on qore
-      branch `bugfix/5348-dpql-infix-word-operator-case` (uncommitted,
-      Nick reinstalls qore himself): lowercased `symbol_to_name` keys +
-      case-insensitive lookup in `resolveExpressionName()`. Validated
-      against pure source (the repo qlib carries a stale compiled
-      `DataProvider.qmod` that qore prefers over source — moved aside for
-      the test, restored after; ~/omq has no qmod, so plain source
-      install works): infix `startsWith`/`STARTSWITH` parse to
-      `starts-with`, function form + `like` unregressed, full
-      serialize→parse round-trip green.
-      **DEPLOYED locally 2026-06-11** (Nick reinstalled qore + restarted
-      core): live `dpql/parse('"test" startsWith "t"')` → success, zero
-      diagnostics; LiveExplain seeded editor clean (no banner/squiggles),
-      Explain still server-rendered. The qore branch stays uncommitted;
-      issue #5348 open until the fix is committed/PR'd upstream.
-- [ ] USER browser click-through per `VERIFY.local.md` § "Render-expression
-      over LSP", incl. IDE :6007 comparison.
-- [ ] `/tmp/render-expression-exposure.md` marked resolved (done — outcome
-      section appended).
-- [ ] Commit only after user approval.

package/.tasks/REQRAFT_AUTO_FIELD.md

@@ -1,176 +0,0 @@
-# Reqraft `auto` field — type-picker + polymorphic dispatch
-
-Phase 1 of the field migration (see
-[`design/FORM_ENGINE_FIELD_MIGRATION.md`](../design/FORM_ENGINE_FIELD_MIGRATION.md)).
-Ports qorus-ide's `AutoField` polymorphic renderer into Reqraft's
-existing `FormField` / `TemplateField` / `FormEngine` stack. This is
-the **keystone** — the schema editor (phase 2) and most of the
-remaining field gap depend on it.
-
-**Status:** done pending user verify — implemented 2026-06-09. 8/8 story
-play tests pass (run via `test-storybook`), prod typecheck + eslint clean,
-verified in an isolated Storybook (picker, type inference, null toggle,
-live dropdown→Integer→number field, and the full FormEngine→…→AutoFormField
-chain). Uncommitted on `feature/dpql-editor` per the STOP-before-commit rule.
-
-**Scope:** one new field component + small dispatcher wiring. Delegates
-all concrete-type rendering back to the existing `FormField` switch;
-does **not** re-implement per-type fields.
-**Estimated size:** ~250–350 lines + 1 story file + 1 test file. ~1–2 days.
-
-## Reference
-
-- Source to port: `qorus-frontend/qorus-ide/src/components/Field/auto.tsx`
-  (type resolution ~`195-268`, type picker ~`933-1014`, soft-types list
-  ~`945`, null toggle ~`308-317`).
-- Reqraft dispatcher: [`src/components/form/fields/Field.tsx`](../src/components/form/fields/Field.tsx)
-  (the `FormField` switch + `renderAllowedValues`).
-- Reqraft type map (two copies to update):
-  `Field.tsx:34-46` and
-  [`TemplateField.tsx:49-77`](../src/components/form/fields/template/TemplateField.tsx).
-- Existing helper to reuse: `getTypeFromValue` from
-  `src/helpers/validations` (already imported by `TemplateField`).
-- Type-picker UI: reuse `SelectFormField`
-  (`src/components/form/fields/select/Select.tsx`).
-
-## Surface area
-
-| File | Change |
-|---|---|
-| `src/components/form/fields/auto/AutoFormField.tsx` | **new** — type-picker, `auto`/`any` resolution, null toggle, delegates concrete render to `FormField` |
-| `src/components/form/fields/auto/AutoFormField.stories.tsx` | **new** — picker, inferred-type, null, allowed-values stories |
-| `src/components/form/fields/Field.tsx` | add `case 'auto': case 'any':` → `AutoFormField`; stop mapping `auto`→`long-string` |
-| `src/components/form/fields/template/TemplateField.tsx` | same map fix so `auto`/`any` no longer collapse to `long-string` |
-| `src/components/form/index.tsx` | export `AutoFormField` |
-| `src/types/Form.ts` | ensure `'auto'` / `'any'` are valid `TFormFieldType` members (they derive from `TQorusType`; confirm) |
-| `__tests__/form/AutoFormField.test.tsx` | **new** — type resolution + picker behaviour |
-
-## Items
-
-### 1. `AutoFormField` component
-
-**Decision.** New component that owns *type selection*, then hands the
-resolved concrete type to the existing `FormField`. It is the Reqraft
-analog of `AutoField` but thin — `FormField` already knows how to
-render every concrete type, so `AutoFormField` must not duplicate that
-switch.
-
-**Behaviour.**
-- Resolve an internal type: `defaultType` if given, else
-  `getTypeFromValue(value)`; `auto`/`any` stay unresolved until the
-  user picks.
-- When unresolved (or `allowedTypes.length > 1`), render the
-  type-picker (item 2), then render the chosen type's field by
-  delegating to `<FormField type={resolved} value onChange ... />`.
-- When resolved, render `<FormField>` directly for that type.
-- Reuse `FormField`'s allowed-values rendering rather than
-  re-implementing it (either render `<FormField>` which already calls
-  `renderAllowedValues`, or lift the helper — prefer the former).
-
-**Props.** Mirror the relevant subset of `IAutoFieldProps`:
-`type` (`'auto'`/`'any'` or a concrete type), `value`, `onChange`,
-`defaultType?`, `allowedTypes?: { name/value }[]`, `canBeNull?`,
-`disabled`, `readonly`, `size`, plus the allowed-values props already
-on `IFormFieldProps`. **Stub but do not wire** `requestFieldData` /
-`type-depends-on` (IDE-only).
-
-**Surface area.** `AutoFormField.tsx` (~150–200 lines).
-
-**Tests.** Unit: given `value=42` + `type='auto'` resolves to a number
-field; given `defaultType='hash'` renders the hash branch; picking a
-type from the picker swaps the rendered sub-field.
-
-### 2. Type picker
-
-**Decision.** A `SelectFormField` of soft types, shown when the type is
-`auto`/`any` or `allowedTypes` has >1 entry. Default soft-type list
-(from auto.tsx ~`945`):
-`bool, softbool, date, string, softstring, binary, float, softfloat,
-list, softlist, hash, int, softint, rgbcolor`. When `allowedTypes` is
-provided, use it verbatim instead of the soft list.
-
-**Note.** Soft types (`softint`, `softstring`, …) resolve to their
-hard counterpart for *rendering* (a `softint` renders the number
-field) but the chosen type is preserved in the emitted value's `type`
-metadata, matching qorus-ide. Confirm `FormField` tolerates a `soft*`
-type by extending `mapQorusTypeToFormFieldType` (soft* → base type).
-
-**Surface area.** Inside `AutoFormField.tsx`; small `mapQorus...`
-extension for `soft*`.
-
-**Tests.** Picker visible only for `auto`/`any`/multi-`allowedTypes`;
-selecting `int` renders the number field.
-
-### 3. Null toggle
-
-**Decision.** When `canBeNull`, render a "Set as null" / "Unset null"
-button (qorus-ide auto.tsx `308-317`). Setting null emits the server's
-null sentinel and hides the field; unsetting restores the prior value.
-
-**Surface area.** `AutoFormField.tsx` (~20 lines).
-
-**Tests.** Toggle sets/clears null; field hidden while null.
-
-### 4. Dispatcher wiring
-
-**Decision.** Add `case 'auto': case 'any':` to `FormField`'s switch
-returning `<AutoFormField ... />`. Fix **both**
-`mapQorusTypeToFormFieldType` copies so `auto`/`any` no longer fall
-through to `'long-string'`. Because `FormEngine` → `TemplateField` →
-`FormField`, schema-driven nested forms gain `auto` with no further
-work.
-
-**Surface area.** `Field.tsx`, `TemplateField.tsx`, `index.tsx` export.
-
-**Tests.** `FormEngine` story with an options schema containing a field
-typed `auto` renders the picker (not a textarea).
-
-## Sequencing
-
-1. Item 1 (component skeleton, delegates to `FormField`)
-2. Item 2 (type picker) — the core of the value
-3. Item 3 (null toggle) — independent, small
-4. Item 4 (dispatcher wiring + map fix) — last, flips the behaviour on
-
-## Tasks
-
-- [x] `AutoFormField.tsx` — type resolution via `getTypeFromValue` /
-      `defaultType`; delegates concrete render to `FormField`
-- [x] Type picker (`SelectFormField`) with soft-type list; `allowedTypes`
-      override; gated on `auto`/`any`/multi-type; honours `noSoft`
-- [x] `soft*` → base-type handling. In `AutoFormField.getRenderType`
-      (`softX`→`X`, `binary`→`long-string`) AND, as a later pass, in
-      `FormField` itself (`renderField` strips the `soft` prefix; `binary`
-      added to the `long-string` case) and in both `mapQorusTypeToFormFieldType`
-      copies. So a bare `<FormField type="softint" />` / `type="binary"` now
-      renders correctly. Stories: `Field.stories.tsx` `SoftInteger` + `Binary`.
-- [x] Null toggle gated on `canBeNull`
-- [x] `case 'auto': case 'any':` in `FormField`; `auto`/`any` now map to
-      `'auto'` in both type maps instead of falling through to `long-string`
-- [x] Export `AutoFormField` from `form/index.tsx`
-- [x] Confirm `'auto'`/`'any'` typecheck as `TFormFieldType`
-      (`TFormFieldValueType` already maps them to `any`)
-- [x] Stories: picker (`Empty`), inferred-from-value (`InferredInteger`,
-      `InferredText`), concrete (`ConcreteString`), null (`Nullable`),
-      allowed-types subset, no-soft, and `FormEngine`-with-auto-field
-      (`ViaFormEngine`)
-- [x] Tests: story play tests cover resolution + null + concrete + inference
-      + the FormEngine chain (run via `test-storybook`, 8/8 pass). No
-      separate jest file — fields are tested via stories in this repo.
-- [x] Prod typecheck, eslint, storybook play tests pass
-- [ ] **STOP — user verifies in browser** before commit
-
-## Out of scope
-
-- ~~Bare `soft*` types passed directly to `FormField`/`TemplateField`.~~
-  **Done** (2026-06-09 follow-up): `FormField.renderField` strips the
-  `soft` prefix, `binary` was added to the `long-string` case, and both
-  `mapQorusTypeToFormFieldType` copies strip `soft`. Verified by
-  `Field.stories.tsx` (`SoftInteger`, `Binary`).
-- `type-depends-on` / `requestFieldData` cross-field resolution (IDE-only)
-- `data-provider` / `connection` / `InterfaceSelector` branches from
-  qorus-ide's AutoField (bucket 3)
-- Expression/template handling beyond what `TemplateField` already does
-- The remaining bucket-1 fields (`byte-size`, `url`, `option_hash`,
-  `code-editor`, …) — separate task files; several get cheaper once
-  this lands