[feat] acknowledge remote input and wait for visual changes

apps/mcp/src/server.ts

@@ -296,6 +296,16 @@ export function createServer(
     },
     { ui: true },
   );
+  add(
+    "crossdesk_wait",
+    z.strictObject({ session_id: id, observation_id: id,
+      condition: z.enum(["changed", "stable"]),
+      stable_ms: z.int().min(100).max(5000).default(500),
+      timeout_ms: z.int().min(0).max(25000).default(5000) }),
+    "Wait for the observed screenshot region to change or remain stable. Returns matched, wait_timed_out and a fresh screenshot on success. A visual condition can be checked after an action with a separate screenshot observation. Remote input receipts report native input acceptance; verify the application's intended result visually.",
+    async (p, ctx) => output(await sessions.waitFor(p, ctx.mcpReq.signal)),
+    { readOnly: true },
+  );
   add(
     "crossdesk_set_control",
     z.strictObject({ session_id: id, enabled: z.boolean() }),

apps/mcp/src/sessions.ts

@@ -459,6 +459,13 @@ export class Sessions {
     }
     return result;
   }
+  async waitFor(p: Data, signal?: AbortSignal): Promise<Data> {
+    const { session_id, ...args } = p;
+    const result = await this.session(session_id).native.request(
+      "wait_frame", args, (args.timeout_ms ?? 5000) + 3000, signal);
+    if (result.matched) result.observation = await this.screenshot({ session_id }, signal);
+    return result;
+  }
   async control(id: string, enabled: boolean) {
     if (!enabled) this.cancelUnlock(this.session(id));
     return this.session(id).native.request("control", { enabled });