fix(linsight): thread task-mode skill selection to the submit

Follow-up to the None≡[] skill gate: with skills now strictly opt-in, the
daily task-mode path (ChatView → useAiChat → unified /chat/completions →
_to_linsight_submit) exposed a latent gap — it never carried the picked
skills, so the stored SV had skills=None and NOTHING was materialized. The
picker looked broken ("workspace has no such skill file"). Previously this
was masked because None loaded every enabled skill.

Thread the selection end-to-end:
- APIChatCompletion gains `skills` (Track H); _to_linsight_submit maps it
  onto the linsight submit schema (None/[] = no skills, opt-in list = those).
- useAiChat sends the picked skill names (taskModeSkillsState('new')) on
  task-mode turns only; the daily chain ignores the field.
- tests: _to_linsight_submit forwards a selection / stays None when absent.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

src/backend/bisheng/api/v1/schema/chat_schema.py

@@ -100,6 +100,13 @@ class APIChatCompletion(BaseModel):
     # daily chain; the turn still lives in the same conversation (chat_id).
     task_mode: bool | None = False
 
+    # --- F035 Track H: per-turn selected skill names (task mode only) ---
+    # The skills the user picked in the daily task-mode input. Threaded onto the
+    # linsight submit schema by ``_to_linsight_submit``; ``None``/``[]`` both mean
+    # "no skills this turn" (skills are opt-in, see materialize_session_skills).
+    # Ignored on the plain daily chain.
+    skills: list[str] | None = None
+
     # --- Preserved (new code-path also honours use_knowledge_base / files) ---
     use_knowledge_base: UseKnowledgeBaseParam | None = None
     files: list[dict] | None = None

src/backend/bisheng/workstation/domain/services/chat_service.py

@@ -2031,4 +2031,7 @@ def _to_linsight_submit(data: APIChatCompletion):
         model=data.model or None,
         files=submit_files,
         tools=submit_tools,
+        # F035 Track H: thread the daily task-mode skill picker onto the submit so
+        # only the user's explicit selection is materialized (None/[] = no skills).
+        skills=data.skills,
     )

src/backend/test/workstation/test_unified_chat_entry.py

@@ -106,3 +106,28 @@ def test_to_linsight_submit_no_knowledge_defaults_false():
 
     assert submit.org_knowledge_enabled is False
     assert submit.personal_knowledge_enabled is False
+
+
+def test_to_linsight_submit_forwards_selected_skills():
+    """The daily task-mode skill picker must reach the submit schema.
+
+    Regression: _to_linsight_submit dropped skills entirely, so the stored SV had
+    skills=None. That was masked while None meant "load all enabled skills"; once
+    None means "no skills", an explicit pick that never arrived silently loaded
+    nothing — the picker looked broken.
+    """
+    data = APIChatCompletion(
+        clientTimestamp="t", conversationId="c", model="m", text="hi", task_mode=True, skills=["morning-report"]
+    )
+
+    submit = chat_service._to_linsight_submit(data)
+
+    assert submit.skills == ["morning-report"]
+
+
+def test_to_linsight_submit_no_skills_stays_none():
+    data = APIChatCompletion(clientTimestamp="t", conversationId="c", model="m", text="hi", task_mode=True)
+
+    submit = chat_service._to_linsight_submit(data)
+
+    assert submit.skills is None

src/frontend/client/src/hooks/useAiChat.ts

@@ -17,7 +17,7 @@ import useAiChatSSE, { type SSESubmission } from "~/hooks/useAiChatSSE";
 import { useGetBsConfig } from "~/hooks/queries/data-provider";
 import { useLinsightManager } from "~/hooks/useLinsightManager";
 import { startLinsight, getLinsightSessionVersionList } from "~/api/linsight";
-import { SopStatus } from "~/store/linsight";
+import { SopStatus, taskModeSkillsState } from "~/store/linsight";
 
 const NO_PARENT = "00000000-0000-0000-0000-000000000000";
 
@@ -46,6 +46,10 @@ export default function useAiChat(initialConversationId: string = "new", isLings
     // emits the new ChatResponse SSE format). Empty array keeps us on the
     // legacy flow so existing tests / old clients aren't disrupted.
     const [selectedAgentTools] = useRecoilState(store.selectedAgentTools);
+    // F035 Track H: skills picked in the daily task-mode input live in the shared
+    // 'new' atom (AiChatInput keys the picker there). Threaded onto the task-mode
+    // turn's submit payload so the user's explicit selection is what gets loaded.
+    const [dailyTaskSkills] = useRecoilState(taskModeSkillsState('new'));
     // Admin-level org-KB toggle. Knowledge spaces remain available even when
     // the org knowledge base feature is disabled, so we only strip org ids.
     const { data: bsConfig } = useGetBsConfig();
@@ -244,6 +248,10 @@ export default function useAiChat(initialConversationId: string = "new", isLings
                 // F035 Track J (TJ-6): route this turn to the linsight task kernel
                 // via the SAME unified entry. Backend replies with a handoff event.
                 task_mode: taskMode,
+                // F035 Track H: send the picked skill names on task-mode turns so
+                // the backend materializes exactly those (empty = none). Omitted
+                // outside task mode (the daily chain ignores it).
+                skills: taskMode ? dailyTaskSkills.map((s) => s.name) : [],
             };
 
             // Correlation key for the user (question) message. Starts as the
@@ -609,7 +617,7 @@ export default function useAiChat(initialConversationId: string = "new", isLings
             setIsStreaming(true);
             setSseSubmission(submission);
         },
-        [conversationId, isStreaming, chatModel, selectedOrgKbs, searchType, selectedAgentTools, isLingsi, createLinsight, updateLinsight, localize, queryClient]
+        [conversationId, isStreaming, chatModel, selectedOrgKbs, searchType, selectedAgentTools, dailyTaskSkills, isLingsi, createLinsight, updateLinsight, localize, queryClient]
     );
 
     // --- Stop generating ---