feat(setup): Yes-default + session-persist on claude-assist, quieter first-chat

Three UX tweaks after watching a user walk through setup:

1. Claude-assist "Run this command?" now defaults to Yes. After Claude has already been asked to diagnose + explained the fix, the vast majority of users want to run it — the No-default added friction without proportional safety.

2. claude-assist persists its session across failures in one setup run. First invocation captures session_id from the stream-json init event; subsequent invocations pass --resume <id>. Claude sees prior failures as conversation history instead of treating each hiccup as a blank-slate ticket.

3. First-chat flow no longer drops the user into a free-text chat loop by default. Instead: explain what the ping/pong check is doing, wait for the pong, then offer "Continue with setup" (recommended, default) or "Pause here and chat with your agent from the terminal" (opt-in). The free-text loop is still reachable, just not the default path.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

setup/auto.ts

@@ -246,10 +246,33 @@ async function main(): Promise<void> {
       );
     }
     if (!skip.has('first-chat')) {
+      p.log.message(
+        dimWrap(
+          "Your assistant runs in an isolated sandbox. I'm going to send it a quick test message (ping) and wait for a reply (pong) to confirm it's responding. First startup typically takes 30–60 seconds while the sandbox warms up.",
+          4,
+        ),
+      );
       const ping = await confirmAssistantResponds();
       if (ping === 'ok') {
         phEmit('first_chat_ready');
-        await runFirstChat();
+        const next = ensureAnswer(
+          await p.select({
+            message: 'What next?',
+            options: [
+              {
+                value: 'continue',
+                label: 'Continue with setup',
+                hint: 'recommended',
+              },
+              {
+                value: 'chat',
+                label: 'Pause here and chat with your agent from the terminal',
+              },
+            ],
+          }),
+        ) as 'continue' | 'chat';
+        setupLog.userInput('first_chat_choice', next);
+        if (next === 'chat') await runFirstChat();
       } else {
         phEmit('first_chat_failed', { reason: ping });
         renderPingFailureNote(ping);

setup/lib/claude-assist.ts

@@ -115,7 +115,7 @@ export async function offerClaudeAssist(
   const run = ensureAnswer(
     await p.confirm({
       message: 'Run this command? (you can edit it before executing)',
-      initialValue: false,
+      initialValue: true,
     }),
   );
   if (!run) return false;
@@ -279,18 +279,24 @@ async function queryClaudeUnderSpinner(
     // No hard timeout — debugging can take a long time, and the cost of
     // cutting Claude off mid-investigation is worse than letting the
     // spinner run. The user can Ctrl-C if they want to abort.
-    const child = spawn(
-      'claude',
-      [
-        '-p',
-        '--output-format',
-        'stream-json',
-        '--verbose',
-        '--permission-mode',
-        'bypassPermissions',
-      ],
-      { cwd: projectRoot, stdio: ['pipe', 'pipe', 'pipe'] },
-    );
+    //
+    // Resume the same session on repeat invocations so Claude carries
+    // context across failures in one setup run.
+    const claudeArgs = [
+      '-p',
+      '--output-format',
+      'stream-json',
+      '--verbose',
+      '--permission-mode',
+      'bypassPermissions',
+    ];
+    if (claudeSessionId) {
+      claudeArgs.push('--resume', claudeSessionId);
+    }
+    const child = spawn('claude', claudeArgs, {
+      cwd: projectRoot,
+      stdio: ['pipe', 'pipe', 'pipe'],
+    });
 
     child.stdout.on('data', (c: Buffer) => {
       lineBuf += c.toString('utf-8');
@@ -301,6 +307,16 @@ async function queryClaudeUnderSpinner(
         if (!line.trim()) continue;
         try {
           const event = JSON.parse(line) as StreamEvent;
+          // Capture the session id on the very first claude invocation of
+          // this process so later calls can --resume it.
+          if (
+            !claudeSessionId &&
+            event.type === 'system' &&
+            event.subtype === 'init' &&
+            typeof event.session_id === 'string'
+          ) {
+            claudeSessionId = event.session_id;
+          }
           handleStreamEvent(event, {
             setAction: (a) => {
               actions.push(a);
@@ -335,10 +351,14 @@ async function queryClaudeUnderSpinner(
 }
 
 // Minimal shape of the stream-json events we care about. Claude emits
-// many more, but we only read tool_use blocks (for breadcrumbs) and text
-// blocks (to reassemble the final REASON/COMMAND answer).
+// many more, but we only read tool_use blocks (for breadcrumbs), text
+// blocks (to reassemble the final REASON/COMMAND answer), and the
+// session_id on the init event so follow-up invocations can resume the
+// same conversation.
 interface StreamEvent {
   type: string;
+  subtype?: string;
+  session_id?: string;
   message?: {
     content?: Array<
       | { type: 'text'; text: string }
@@ -347,6 +367,11 @@ interface StreamEvent {
   };
 }
 
+// The session id from the first claude-assist invocation in this process.
+// Subsequent invocations pass `--resume <id>` so Claude sees prior failures
+// as conversation history instead of treating each failure in isolation.
+let claudeSessionId: string | null = null;
+
 function handleStreamEvent(
   event: StreamEvent,
   cb: { setAction: (a: string) => void; appendText: (t: string) => void },