feat(agent): unify MCP tools and adopt Responses delivery

Move project and host queries into native MCP with host-owned OAuth/context,
while retaining memory, skills, actions and publication plugins. Add nearby-model
queries and explicit chat/task/learning model defaults.

Adopt Responses HTTP/SSE and native steer/queue inbox delivery, preserving scoped
session ownership, pending input edits and investigation communication.

Disable local-provider discovery: Server /health allowed native vLLM to probe
/v1/models on port 8000. A native before/after regression prevents recurrence.

Validation: 170 focused Bun tests; service/plugin typechecks; native discovery,
provider-discovery and input-inbox smoke checks; no paid model inference.

src/routes/chatAuxiliaryRoutes.ts

@@ -4,12 +4,9 @@ import { z } from "zod";
 import { getLocalAgentContext } from "../context/localContext.js";
 import { type ChatSessionBridge } from "../chat/sessionBridge.js";
 import { logger } from "../logger.js";
-import { type ResultReferenceResolver } from "../results/resolver.js";
-import { RESULT_REFERENCE_KIND } from "../results/store.js";
 import { type SessionMetadataStore } from "../sessions/metadataStore.js";
 import { type SessionUiStateStore } from "../sessions/uiStateStore.js";
 import { type FrontendActionCoordinator } from "../frontendAction/coordinator.js";
-import { toActorKey, toProjectKey } from "../utils/fileStore.js";
 import {
   type ActiveRun,
   type RunStatus,
@@ -25,15 +22,16 @@ const abortPayloadSchema = z.object({
 type RegisterAuxiliaryRoutesOptions = {
   activeRuns: Map<string, ActiveRun>;
   lastRunStatuses: Map<string, RunStatus>;
-  resultReferenceResolver: ResultReferenceResolver;
   sessionBridge: ChatSessionBridge;
   sessionMetadataStore: SessionMetadataStore;
   sessionUiStateStore: SessionUiStateStore;
   frontendActionCoordinator: FrontendActionCoordinator;
 };
 
-const toSessionUiStateContext = (sessionId: string) => ({
-  sessionId,
+const toSessionUiStateContext = (record: { sessionId: string; actorKey: string; projectKey: string }) => ({
+  sessionId: record.sessionId,
+  actorKey: record.actorKey,
+  projectKey: record.projectKey,
 });
 
 export const registerChatAuxiliaryRoutes = (
@@ -41,38 +39,12 @@ export const registerChatAuxiliaryRoutes = (
   {
     activeRuns,
     lastRunStatuses,
-    resultReferenceResolver,
     sessionBridge,
     sessionMetadataStore,
     sessionUiStateStore,
     frontendActionCoordinator,
   }: RegisterAuxiliaryRoutesOptions,
 ) => {
-  chatRouter.get("/result-ref/:result_ref", async (req, res) => {
-    const resultRef = req.params.result_ref?.trim();
-    const context = getLocalAgentContext(req);
-    const clientSessionId = typeof req.query.session_id === "string" ? req.query.session_id.trim() : undefined;
-    const offset = Math.max(0, Number.parseInt(String(req.query.offset ?? "0"), 10) || 0);
-    const limit = Math.min(500, Math.max(1, Number.parseInt(String(req.query.limit ?? "100"), 10) || 100));
-    if (!resultRef || !clientSessionId) {
-      res.status(400).json({ message: "result_ref and session_id are required" });
-      return;
-    }
-    const result = await resultReferenceResolver.getFullAuthorized(resultRef, {
-      actorKey: context.actorKey, clientSessionId, projectId: context.projectId,
-    }, { expectedKind: RESULT_REFERENCE_KIND.serverApiPayload });
-    if (!result) {
-      res.status(404).json({ message: "result_ref not found" });
-      return;
-    }
-    const rows = Array.isArray(result.data)
-      ? result.data
-      : result.data && typeof result.data === "object" && Array.isArray((result.data as { data?: unknown }).data)
-        ? (result.data as { data: unknown[] }).data
-        : [result.data];
-    res.json({ ...result, data: rows.slice(offset, offset + limit), page: { offset, limit, total: rows.length, has_more: offset + limit < rows.length } });
-  });
-
   chatRouter.post("/abort", async (req, res) => {
     const parsed = abortPayloadSchema.safeParse(req.body);
     if (!parsed.success) {
@@ -98,6 +70,7 @@ export const registerChatAuxiliaryRoutes = (
         return;
       }
       frontendActionCoordinator.cancelSession(sessionRecord.sessionId);
+      activeRuns.get(sessionRecord.sessionId)?.inputs?.close();
       const binding = await sessionBridge.abort({
         clientSessionId: sessionRecord.sessionId,
         sessionId: sessionRecord.sessionId,
@@ -118,9 +91,9 @@ export const registerChatAuxiliaryRoutes = (
           todos: cancelBackendTodos(message.todos),
         }));
         const currentState = await sessionUiStateStore.read(
-          toSessionUiStateContext(sessionRecord.sessionId),
+          toSessionUiStateContext(sessionRecord),
         );
-        await sessionUiStateStore.write(toSessionUiStateContext(sessionRecord.sessionId), {
+        await sessionUiStateStore.write(toSessionUiStateContext(sessionRecord), {
           sessionId: sessionRecord.sessionId,
           isTitleManuallyEdited: currentState?.isTitleManuallyEdited ?? false,
           messages: run.messages,
@@ -130,9 +103,9 @@ export const registerChatAuxiliaryRoutes = (
             session_id: parsed.data.session_id,
             message: "请求已中断",
           });
-          subscriber.close();
         }
-        run.subscribers.clear();
+        // The generation route emits its terminal Response after cleanup and
+        // closes subscribers. Closing here would truncate the Responses stream.
       }
 
       logger.info(