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/product/task-presentation.ts

@@ -0,0 +1,46 @@
+import { z } from "zod";
+
+import { safeChartInput } from "../frontendAction/registry.js";
+
+export const supplyLayerSchema = z.enum(["junctions", "pipes", "valves", "pumps", "tanks", "reservoirs", "scada"]);
+const identifier = z.string().trim().min(1).max(128);
+
+export const taskPresentationCommandSchema = z.discriminatedUnion("action", [
+  z.object({
+    action: z.literal("highlight_features"),
+    layer: supplyLayerSchema,
+    feature_ids: z.array(identifier).min(1).max(100),
+    effect: z.enum(["steady", "pulse"]).default("steady"),
+  }).strict(),
+  z.object({ action: z.literal("clear_highlights") }).strict(),
+  z.object({
+    action: z.literal("open_chart"),
+    chart_id: identifier,
+    chart: safeChartInput,
+    evidence_refs: z.array(z.string().trim().min(1)).min(1).max(100),
+  }).strict(),
+  z.object({ action: z.literal("close_chart"), chart_id: identifier }).strict(),
+  z.object({
+    action: z.literal("set_overlay"),
+    overlay_id: identifier,
+    title: z.string().trim().min(1).max(200),
+    mode: z.enum(["geometry", "heatmap"]),
+    evidence_id: z.string().trim().min(1),
+  }).strict(),
+  z.object({ action: z.literal("clear_overlay"), overlay_id: identifier }).strict(),
+]);
+
+export type TaskPresentationCommand = z.infer<typeof taskPresentationCommandSchema>;
+
+export const taskPresentationResultSchema = z.object({
+  recorded: z.literal(true),
+  event_seq: z.number().int().positive(),
+  run_id: z.string().uuid(),
+  execution_generation: z.number().int().positive(),
+}).strict();
+
+export const taskPresentationBridgeSchema = z.object({
+  session_id: z.string().trim().min(1).max(128),
+  request_id: z.string().uuid(),
+  command: taskPresentationCommandSchema,
+}).strict();