feat(skill): add draft confirmation card with view/install/discard

docs/features/skill-draft-confirmation-card/plan.md

@@ -0,0 +1,46 @@
+# Plan
+
+## Approach
+
+1. Extend skill presenter with draft action helpers that can read, install, and delete a draft by conversation id and draft id.
+2. When `skill_manage` `create` succeeds, include structured metadata in the tool raw result.
+3. In agent runtime dispatch, detect successful draft creation and append a synthetic `question_request` action block using the existing question interaction flow.
+4. Handle the synthetic interaction in `respondToolInteraction`:
+   - View: read draft content and keep the same card pending with content attached.
+   - Install: install draft folder into the configured skills directory, resolve the card, and update the `skill_manage` tool response.
+   - Discard: delete draft, resolve the card, and update the `skill_manage` tool response.
+5. Update renderer question panel to render draft preview content and localized option labels.
+6. Add/update tests for presenter draft actions, tool metadata, runtime interaction behavior, and renderer panel display.
+
+## Affected Interfaces
+
+- `SkillManageResult` gains optional `draftStatus` and supporting metadata only where useful.
+- `ISkillPresenter` gains draft action methods scoped by conversation id/draft id.
+- `MCPToolResponse.toolResult` carries structured draft result metadata for runtime-only detection.
+- Question action block `extra` gains skill-draft-specific metadata fields.
+
+## Data Flow
+
+```text
+Agent skill_manage create
+  -> SkillPresenter.manageDraftSkill creates temp draft
+  -> AgentToolManager returns rawData.toolResult
+  -> dispatch detects draft metadata
+  -> append question_request card
+  -> renderer shows ChatToolInteractionOverlay
+  -> user chooses option
+  -> AgentRuntimePresenter handles draft action
+  -> SkillPresenter view/install/delete draft
+  -> update card/tool response and resume as needed
+```
+
+## Compatibility
+
+Existing question and permission blocks remain unchanged. Draft-specific behavior is guarded by `extra.skillDraftAction === 'confirm'` and the `skill_manage` tool name.
+
+## Test Strategy
+
+- Unit test draft action helpers in `skillPresenter`.
+- Unit test `AgentToolManager` raw metadata for `skill_manage create`.
+- Unit/integration test runtime draft confirmation interactions.
+- Renderer component test for draft preview rendering in `ChatToolInteractionOverlay`.

docs/features/skill-draft-confirmation-card/spec.md

@@ -0,0 +1,41 @@
+# Skill Draft Confirmation Card
+
+## User Need
+
+When the Agent creates a reusable skill draft after a task, users need an obvious in-chat confirmation card instead of having the draft disappear into a temp directory with no visible follow-up path.
+
+## Goal
+
+After a successful `skill_manage` draft creation, show a blocking question-style card in the chat:
+
+- “已生成 skill draft：xxx”
+- Options: 查看内容 / 安装为 Skill / 丢弃
+
+The card should reuse the existing question interaction panel so it fits the current Agent interaction flow and pauses until the user chooses an action.
+
+## Acceptance Criteria
+
+1. A successful `skill_manage` `create` result produces an in-chat question interaction card before the Agent continues.
+2. The card has three options: view content, install as Skill, discard.
+3. Choosing “查看内容” shows the draft `SKILL.md` content and keeps the draft available for a later install/discard choice.
+4. Choosing “安装为 Skill” installs the draft into the configured skills directory and resumes the Agent with a clear success/failure tool result.
+5. Choosing “丢弃” deletes the draft and resumes the Agent with a clear result.
+6. Existing `deepchat_question` interactions continue to work unchanged.
+7. Draft install/delete operations remain scoped to the current conversation and opaque draft id.
+
+## Constraints
+
+- Follow existing presenter boundaries and typed route/contracts patterns where needed.
+- Keep the UI change focused by reusing `ChatToolInteractionOverlay` and `question_request` blocks.
+- Avoid exposing absolute temp paths to the renderer or the model.
+- All user-facing strings must use i18n keys.
+
+## Non-goals
+
+- No full Drafts management list/page in Settings.
+- No long-term draft persistence beyond the current temp draft retention policy.
+- No automatic install without user confirmation.
+
+## Open Questions
+
+Resolved: Use the existing question panel and show a confirmation card immediately after draft creation.

docs/features/skill-draft-confirmation-card/tasks.md

@@ -0,0 +1,10 @@
+# Tasks
+
+- [x] Add draft view/install/delete helpers to SkillPresenter.
+- [x] Add structured draft metadata to skill_manage tool results.
+- [x] Append synthetic draft confirmation question after successful draft create.
+- [x] Handle view/install/discard responses in AgentRuntimePresenter.
+- [x] Render draft preview content in question panel.
+- [x] Add i18n strings.
+- [x] Add/update tests.
+- [ ] Run pnpm run format, pnpm run i18n, pnpm run lint.

src/main/presenter/agentRuntimePresenter/dispatch.ts

@@ -73,9 +73,23 @@ type StagedToolResult = {
   rtkMode?: 'rewrite' | 'direct' | 'bypass'
   rtkFallbackReason?: string
   imagePreviews?: ToolCallImagePreview[]
+  skillDraftPrompt?: SkillDraftPromptPayload
   postHookKind: 'success' | 'failure'
 }
 
+type SkillDraftPromptPayload = {
+  draftId: string
+  skillName: string
+}
+
+type SkillDraftToolResult = {
+  skillDraft?: {
+    status?: unknown
+    draftId?: unknown
+    skillName?: unknown
+  }
+}
+
 type ToolRunOutcome =
   | {
       kind: 'staged'
@@ -399,6 +413,28 @@ function extractSubagentToolState(rawData: MCPToolResponse): {
   }
 }
 
+function extractSkillDraftPromptPayload(
+  rawData: MCPToolResponse
+): SkillDraftPromptPayload | undefined {
+  const toolResult = rawData.toolResult as SkillDraftToolResult | null | undefined
+  const skillDraft = toolResult?.skillDraft
+  if (!skillDraft || typeof skillDraft !== 'object') {
+    return undefined
+  }
+  if (skillDraft.status !== 'created') {
+    return undefined
+  }
+  if (typeof skillDraft.draftId !== 'string' || typeof skillDraft.skillName !== 'string') {
+    return undefined
+  }
+  const draftId = skillDraft.draftId.trim()
+  const skillName = skillDraft.skillName.trim()
+  if (!draftId || !skillName) {
+    return undefined
+  }
+  return { draftId, skillName }
+}
+
 function shouldRefreshToolsAfterCall(toolName: string, rawData: MCPToolResponse): boolean {
   if (toolName !== 'skill_view') {
     return false
@@ -598,6 +634,21 @@ function applyFinalizedToolResults(params: {
       imagePresentation.promotedBlocks
     )
 
+    if (stagedResult.skillDraftPrompt && !fittedResult.isError && !fittedResult.downgraded) {
+      const interaction = appendSkillDraftQuestionActionBlock(
+        state,
+        io,
+        {
+          id: stagedResult.toolCallId,
+          name: stagedResult.toolName,
+          args: stagedResult.toolArgs,
+          serverName: stagedResult.serverName
+        },
+        stagedResult.skillDraftPrompt
+      )
+      state.pendingInteractions = [...(state.pendingInteractions ?? []), interaction]
+    }
+
     if (fittedResult.isError) {
       hooks?.onPostToolUseFailure?.({
         callId: stagedResult.toolCallId,
@@ -733,7 +784,8 @@ function appendQuestionActionBlock(
     serverIcons?: string
     serverDescription?: string
   },
-  question: NonNullable<PendingToolInteraction['question']>
+  question: NonNullable<PendingToolInteraction['question']>,
+  extra?: Record<string, unknown>
 ): PendingToolInteraction {
   state.blocks.push({
     type: 'action',
@@ -756,7 +808,8 @@ function appendQuestionActionBlock(
       questionOptions: question.options,
       questionMultiple: question.multiple,
       questionCustom: question.custom,
-      questionResolution: 'asked'
+      questionResolution: 'asked',
+      ...extra
     }
   })
   state.dirty = true
@@ -773,6 +826,47 @@ function appendQuestionActionBlock(
   }
 }
 
+function buildSkillDraftQuestion(
+  _payload: SkillDraftPromptPayload
+): NonNullable<PendingToolInteraction['question']> {
+  return {
+    header: 'chat.skillDraft.confirmationTitle',
+    question: 'chat.skillDraft.confirmationQuestion',
+    options: [
+      {
+        label: 'chat.skillDraft.actions.view',
+        description: 'chat.skillDraft.actions.viewDescription'
+      },
+      {
+        label: 'chat.skillDraft.actions.install',
+        description: 'chat.skillDraft.actions.installDescription'
+      },
+      {
+        label: 'chat.skillDraft.actions.discard',
+        description: 'chat.skillDraft.actions.discardDescription'
+      }
+    ],
+    custom: false,
+    multiple: false
+  }
+}
+
+function appendSkillDraftQuestionActionBlock(
+  state: StreamState,
+  io: IoParams,
+  toolContext: ToolExecutionContext['toolContext'],
+  payload: SkillDraftPromptPayload
+): PendingToolInteraction {
+  const question = buildSkillDraftQuestion(payload)
+  return appendQuestionActionBlock(state, io, toolContext, question, {
+    skillDraftAction: 'confirm',
+    skillDraftId: payload.draftId,
+    skillDraftName: payload.skillName,
+    skillDraftPreview: '',
+    skillDraftStatus: 'pending'
+  })
+}
+
 function flushBlocksToRenderer(io: IoParams, blocks: AssistantMessageBlock[]): void {
   const renderedBlocks = cloneBlocksForRenderer(blocks)
   eventBus.sendToRenderer(STREAM_EVENTS.RESPONSE, SendTarget.ALL_WINDOWS, {
@@ -961,6 +1055,7 @@ async function runToolCall(params: {
         rtkMode: toolRawData.rtkMode,
         rtkFallbackReason: toolRawData.rtkFallbackReason,
         imagePreviews,
+        skillDraftPrompt: extractSkillDraftPromptPayload(toolRawData),
         postHookKind: stagedIsError ? 'failure' : 'success'
       },
       toolsChanged: shouldRefreshToolsAfterCall(completedToolCall.name, toolRawData)
@@ -995,6 +1090,10 @@ export async function executeTools(
   finalizePendingNarrativeBeforeToolExecution(state)
   persistToolExecutionState(io, state, rendererFlushHandle)
 
+  if (state.pendingInteractions?.length) {
+    state.pendingInteractions = []
+  }
+
   for (const tc of state.completedToolCalls) {
     const toolDef = tools.find((t) => t.function.name === tc.name)
     if (!toolDef) continue
@@ -1155,6 +1254,10 @@ export async function executeTools(
         hooks,
         appendToConversation: fittedResults.kind === 'ok'
       })
+      if (state.pendingInteractions?.length) {
+        pendingInteractions.push(...state.pendingInteractions)
+        state.pendingInteractions = []
+      }
       persistToolExecutionState(io, state, rendererFlushHandle)
 
       if (fittedResults.kind === 'terminal_error') {
@@ -1316,6 +1419,10 @@ export async function executeTools(
       hooks,
       appendToConversation: fittedResults.kind === 'ok'
     })
+    if (state.pendingInteractions?.length) {
+      pendingInteractions.push(...state.pendingInteractions)
+      state.pendingInteractions = []
+    }
     persistToolExecutionState(io, state, rendererFlushHandle)
 
     if (fittedResults.kind === 'terminal_error') {