feat(chat): queue repeated messages from trailing xN intent

docs/superpowers/specs/2026-07-15-repeat-intent-detection-design.md

@@ -0,0 +1,243 @@
+# Repeat Intent Detection Design
+
+**Date:** 2026-07-15  
+**Status:** Approved for implementation planning  
+**Scope:** Chat composer send path + pending message queue (frontend only)  
+**Approach:** Pure trailing-multiplier parser + confirm dialog + bulk enqueue
+
+## Problem
+
+Users often want to queue the same short follow-up many times (for example `继续` / `continue`) while an agent is working through a multi-step task. Typing or clicking send repeatedly is tedious. A common natural shorthand is:
+
+```text
+继续x10
+继续 X10
+continue x 5
+```
+
+Codeg already has a pending-message queue (`useMessageQueue`) and auto-flush when a turn becomes free. What is missing is recognition of the trailing `xN` / `XN` multiplier and a confirmation step that expands it into N queued drafts.
+
+## Goals
+
+1. Detect a trailing repeat multiplier on composer submit.
+2. Support `x` and `X`, with optional spaces around the multiplier.
+3. Match only a **trailing** suffix (not whole-string-only, not mid-text).
+4. Show a confirmation dialog before expanding.
+5. On confirm, enqueue **N identical base messages** into the existing pending queue.
+6. On cancel/dismiss, keep the original composer text unchanged.
+7. Cap N to a safe range: **2..50**.
+8. Cover the behavior with unit tests and i18n (EN + zh-CN minimum; all locales if key parity is enforced).
+
+## Non-goals
+
+- Live detection while typing / inline chips.
+- Direct multi-send that bypasses the queue.
+- Multipliers other than `x` / `X` (`*10`, `×10`, `x10次`, etc.).
+- Persisted "always expand" preference.
+- Backend/Rust changes.
+- Changing auto-flush timing or queue FIFO semantics.
+
+## Decisions (from product clarification)
+
+| Topic | Decision |
+| --- | --- |
+| Match scope | Trailing multiplier suffix only |
+| Confirm action | Enqueue all N items into pending queue |
+| Cancel action | Close dialog only; keep original text including `xN` |
+| Valid N | Integer in **2..50** inclusive |
+| Idle vs busy | Always enqueue N; do not special-case direct first send |
+
+## Detection rules
+
+### Input source
+
+On send, parse the draft's plain `displayText` after normal draft construction.
+
+Do **not** run detection when:
+
+- composer is in queue-item edit mode
+- `buildDraft()` fails / returns empty
+- `onEnqueue` / bulk-enqueue path is unavailable (fall back to normal single send of the original draft)
+
+### Pattern
+
+Conceptual regex against the full display text:
+
+```text
+^(?s)(.+?)[ \t]*[xX][ \t]*(\d+)[ \t]*$
+```
+
+Rules:
+
+1. Trim only trailing whitespace of the whole input before matching if needed; keep internal spaces inside the base text.
+2. `baseText` = capture group 1, right-trimmed; must be non-empty.
+3. `count` = integer from capture group 2.
+4. Trigger only when `2 <= count <= 50`.
+5. Multiplier must be trailing; anything after the number fails the match.
+
+### Positive examples
+
+| Input | baseText | count |
+| --- | --- | --- |
+| `继续X10` | `继续` | 10 |
+| `继续x10` | `继续` | 10 |
+| `继续 X10` | `继续` | 10 |
+| `继续 x 10` | `继续` | 10 |
+| `continue X 3` | `continue` | 3 |
+| `请继续修复x10` | `请继续修复` | 10 |
+| `fix this\nplease x2` | `fix this\nplease` | 2 |
+
+### Negative examples
+
+| Input | Reason |
+| --- | --- |
+| `继续x1` | below min |
+| `继续x51` | above max |
+| `x10` | empty base text |
+| `继续X10 请` | multiplier not trailing |
+| `继续X10a` | non-digit tail |
+| `继续*10` | unsupported operator |
+| attachment-only empty text | no textual multiplier |
+
+### Rich drafts / attachments
+
+- Multiplier is detected only from trailing **text**.
+- On confirm, strip the multiplier from text (`displayText` and text blocks).
+- Keep image/resource/reference blocks unchanged.
+- If stripping leaves empty prose but attachments remain, the resulting draft is still valid (same as normal attachment-only drafts).
+
+## Interaction flow
+
+```text
+User presses Send / Enter
+        |
+        v
+buildDraft()
+        |
+        +-- queue edit mode? --> save edit (no repeat detect)
+        |
+        v
+parseRepeatIntent(displayText)
+        |
+        +-- no match / out of range --> existing send or single enqueue
+        |
+        v
+open AlertDialog (composer text kept)
+        |
+        +-- Cancel / Esc / dismiss --> close only
+        |
+        v
+Confirm
+  - build base draft (suffix stripped)
+  - enqueueMany(baseDraft, modeId, count)
+  - clear composer
+  - existing queue UI + auto-flush handle delivery
+```
+
+### Dialog copy
+
+i18n keys under the chat/message-input namespace:
+
+- Title: "Queue repeated messages?" / "生成重复待发送消息？"
+- Description: include `count` and a short preview of `baseText`
+- Confirm: "Queue N messages" / "生成 N 条待发送"
+- Cancel: reuse existing cancel wording where possible
+
+### Enqueue semantics
+
+- Always enqueue **N** copies of the base draft.
+- Do **not** direct-send the first item even when the agent is idle.
+- Prefer one atomic queue commit (`enqueueMany`) to avoid N React state updates and intermediate renders.
+- Each queued item gets its own id (existing `randomUUID()` behavior).
+- Mode id uses the same value the normal enqueue path would use.
+
+If bulk enqueue plumbing is unavailable on a surface, fall back to normal single send of the **original** unsplit draft. Never silently multi-send outside the queue.
+
+## Architecture
+
+### Components
+
+1. **`src/lib/repeat-intent.ts`** (pure)
+   - `MIN_REPEAT_COUNT = 2`
+   - `MAX_REPEAT_COUNT = 50`
+   - `parseRepeatIntent(text: string): { baseText: string; count: number } | null`
+   - helper(s) to strip the trailing multiplier from draft text/blocks
+
+2. **`src/lib/repeat-intent.test.ts`**
+   - spaces, case, bounds, unicode base text, newlines, non-matches
+
+3. **`src/hooks/use-message-queue.ts`**
+   - add `enqueueMany(draft, modeId, count)`
+   - single `commit([...queue, ...newItems])`
+   - unit coverage for atomic append + count clamp/guard (`count < 1` no-op)
+
+4. **`src/components/chat/message-input.tsx`**
+   - intercept in `handleSend` before normal send/enqueue
+   - local dialog state: open flag + pending `{ draft, modeId, baseText, count }`
+   - render shared `AlertDialog` on confirm/cancel
+
+5. **Parent wiring**
+   - `conversation-detail-panel` already owns `useMessageQueue`
+   - pass bulk enqueue down through `chat-input` / `conversation-shell` as `onEnqueueMany` (or equivalent)
+   - keep existing `onEnqueue` for single-item paths
+
+6. **i18n**
+   - `en.json`, `zh-CN.json`, and any other locales required by key-parity tests
+
+### Data flow
+
+```text
+MessageInput.handleSend
+  -> parseRepeatIntent
+  -> (confirm)
+  -> onEnqueueMany(baseDraft, modeId, count)
+  -> useMessageQueue.enqueueMany
+  -> MessageQueueDisplay + existing auto-flush
+```
+
+### Error handling
+
+- Invalid/out-of-range count: treat as normal message, no dialog.
+- Dialog open while status changes: cancel keeps text; confirm still only enqueues (safe if agent becomes idle/busy).
+- No new backend error paths.
+
+## Testing plan
+
+1. **Parser unit tests**
+   - all positive/negative examples above
+   - leading/trailing spaces around `x` and digits
+   - multiline base text with trailing multiplier
+
+2. **Queue unit tests**
+   - `enqueueMany` appends N items with identical draft/modeId and unique ids
+   - does not mutate previous queue order
+   - `count <= 0` is a no-op
+
+3. **Composer behavior tests** (message-input or focused helper tests)
+   - matching text opens dialog and does not enqueue yet
+   - confirm enqueues N and clears composer
+   - cancel leaves original text
+   - queue-edit mode skips detection
+   - out-of-range sends/enqueues once through normal path
+
+## Implementation notes
+
+- Prefer pure helpers over embedding regex in the React component.
+- Reuse existing `AlertDialog` primitives (`src/components/ui/alert-dialog.tsx`).
+- Keep the feature frontend-only; no Tauri/command changes.
+- Follow Prettier/ESLint project style (no semicolons, etc.).
+- Work from a fresh branch off `main`, not the current Tailscale funnel branch.
+
+## PR expectations
+
+- Feature branch: `feat/repeat-intent-queue` (or similar)
+- PR body includes **English + 中文** summary, behavior, test plan, and screenshots/GIFs if UI is easy to capture
+- Target repository: `xintaofei/codeg` (use fork remote if origin push is unavailable)
+
+## Open questions
+
+None remaining for v1. Future optional enhancements (out of scope now):
+
+- remember "don't ask again" for the session
+- custom max count in settings
+- support for `×` / `*` multipliers

src/components/chat/chat-input.tsx

@@ -46,6 +46,11 @@ interface ChatInputProps {
   showActiveFlow?: boolean
   queue?: QueuedMessage[]
   onEnqueue?: (draft: PromptDraft, modeId: string | null) => void
+  onEnqueueMany?: (
+    draft: PromptDraft,
+    modeId: string | null,
+    count: number
+  ) => void
   onQueueReorder?: (items: QueuedMessage[]) => void
   onQueueEdit?: (id: string) => void
   onQueueDelete?: (id: string) => void
@@ -101,6 +106,7 @@ export const ChatInput = memo(function ChatInput({
   showActiveFlow,
   queue,
   onEnqueue,
+  onEnqueueMany,
   onQueueReorder,
   onQueueEdit,
   onQueueDelete,
@@ -168,6 +174,7 @@ export const ChatInput = memo(function ChatInput({
         isActive={isActive}
         showActiveFlow={showActiveFlow}
         onEnqueue={onEnqueue}
+        onEnqueueMany={onEnqueueMany}
         editingItemId={editingItemId}
         editingDraftText={editingDraftText}
         editingDraftBlocks={editingDraftBlocks}

src/components/chat/conversation-shell.tsx

@@ -73,6 +73,11 @@ interface ConversationShellProps {
   showActiveFlow?: boolean
   queue?: QueuedMessage[]
   onEnqueue?: (draft: PromptDraft, modeId: string | null) => void
+  onEnqueueMany?: (
+    draft: PromptDraft,
+    modeId: string | null,
+    count: number
+  ) => void
   onQueueReorder?: (items: QueuedMessage[]) => void
   onQueueEdit?: (id: string) => void
   onQueueDelete?: (id: string) => void
@@ -126,6 +131,7 @@ export function ConversationShell({
   showActiveFlow,
   queue,
   onEnqueue,
+  onEnqueueMany,
   onQueueReorder,
   onQueueEdit,
   onQueueDelete,
@@ -250,6 +256,7 @@ export function ConversationShell({
               showActiveFlow={showActiveFlow}
               queue={queue}
               onEnqueue={onEnqueue}
+              onEnqueueMany={onEnqueueMany}
               onQueueReorder={onQueueReorder}
               onQueueEdit={onQueueEdit}
               onQueueDelete={onQueueDelete}