docs: clarify fanout_dispatch guideline about auto delivery vs collect

- Include job output preview as indented code block in fanout_status results
- Preview text is already available in status().jobs[].preview via controller
- Now displayed in the text content output alongside job metadata
- Each preview line is indented for readability in markdown code blocks
- Useful for quick inspection of job results without calling fanout_collect

controller.ts

@@ -245,7 +245,8 @@ export class FanoutController {
   }
 
   private getJobPreview(job: FanoutJob, maxChars = 500): string {
-    const output = this.readJobOutput(job);
+    // Use cached preview if available (updated in processOutputLine)
+    const output = job.lastPreview || this.readJobOutput(job);
     if (!output) return "";
     if (output.length <= maxChars) return output;
     return output.slice(0, maxChars) + "\n… (truncated)";
@@ -257,6 +258,7 @@ export class FanoutController {
     if (event.type === "message_end" && event.message) {
       const msg = event.message as Message;
       if (msg.role === "assistant") {
+        // Cache latest assistant text for live preview
         job.usage.turns++;
         const usage = msg.usage;
         if (usage) {
@@ -269,6 +271,21 @@ export class FanoutController {
         if (!job.modelUsed && msg.model) job.modelUsed = msg.model;
         if (msg.stopReason) job.stopReason = msg.stopReason;
         if (msg.errorMessage) job.errorMessage = msg.errorMessage;
+        
+        // Cache last assistant text output for live preview
+        let previewUpdated = false;
+        for (const part of msg.content) {
+          if (part.type === "text") {
+            job.lastPreview = part.text;
+            previewUpdated = true;
+            break;
+          }
+        }
+        
+        // Persist cache to enable live previews after restart
+        if (previewUpdated) {
+          this.persist(job);
+        }
       }
     }
   }

index.ts

@@ -51,7 +51,7 @@ export default function (pi: ExtensionAPI) {
     promptGuidelines: [
       "Use fanout_dispatch to run multiple agents in parallel without blocking the main session.",
       "After dispatching, you may do other work. Check status later with fanout_status.",
-      "When jobs complete, the system will send a follow-up message. Retrieve output with fanout_collect.",
+      "Completed jobs deliver their output automatically via follow-up messages. Use fanout_collect if you need the full, untruncated result or to archive old jobs.",
     ],
 
     execute: async (
@@ -126,12 +126,14 @@ export default function (pi: ExtensionAPI) {
         result.jobs.length > 0
           ? "\n\n" +
             result.jobs
-              .map(
-                (j) =>
+              .map((j) => {
+                let jobLine =
                   `- ${j.id} | ${j.agent} | ${j.status}` +
                   (j.exitCode !== undefined ? ` (exit ${j.exitCode})` : "") +
-                  (j.turns ? ` | ${j.turns} turns` : ""),
-              )
+                  (j.turns ? ` | ${j.turns} turns` : "") +
+                  (j.preview ? `\n  Preview: ${j.preview.slice(0, 120).replace(/\n/g, " ")}${j.preview.length > 120 ? "…" : ""}` : "");
+                return jobLine;
+              })
               .join("\n")
           : "";
       return {

types.ts

@@ -26,6 +26,8 @@ export interface FanoutJob {
   usage: JobUsage;
   modelUsed?: string;
   stopReason?: string;
+  /** Last assistant text output seen while running (for live previews) */
+  lastPreview?: string;
 }
 
 export interface DispatchResult {