feat: chat uploads, vision turn isolation, and MindSpace agent improvements

Add chat file/image upload UX, attachment proxying, vision thumbnails, and per-turn image scoping so agents only use the current upload. Extend MindSpace asset context, billing token state, OA/scenario verify scripts, and related runtime config.

Co-authored-by: Cursor <cursoragent@cursor.com>

mindspace-agent-runner.mjs

@@ -5,6 +5,7 @@ import { Agent, fetch as undiciFetch } from 'undici';
 import { jsonrepair } from 'jsonrepair';
 import { reconcileAgentSession } from './session-reconcile.mjs';
 import { resolveSessionAccess } from './session-broker.mjs';
+import { resolveBillingTokenState } from './billing-token-state.mjs';
 
 const insecureDispatcher = new Agent({
   connect: { rejectUnauthorized: false },
@@ -395,7 +396,15 @@ export function createMindSpaceAgentRunner({
       const parsed = normalizeStructuredResult(extractJsonObject(reply.text));
 
       if (reply.tokenState) {
-        await userAuth.billSessionUsage(claim.userId, sessionId, reply.tokenState, requestId);
+        const tokenState = await resolveBillingTokenState(reply.tokenState, {
+          sessionId,
+          fetchSession: async (sid) => {
+            const response = await apiFetch(`/sessions/${encodeURIComponent(sid)}`);
+            if (!response?.ok) return null;
+            return readJsonResponse(response);
+          },
+        });
+        await userAuth.billSessionUsage(claim.userId, sessionId, tokenState, requestId);
       }
 
       const completed = await agentJobService.completeJob(jobId, claim.jobToken, {