From 21fca7642c1a0ae08ed0a5f3650ce5453bfa5cae Mon Sep 17 00:00:00 2001
From: Sweets Sweetman <sweetmantech@gmail.com>
Date: Thu, 19 Mar 2026 18:32:58 -0500
Subject: [PATCH] fix: add debug logging for sandbox runCommand 400 errors

Logs sandboxId, cwd, command details when runCommand fails
to help diagnose transient Vercel Sandbox API errors.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---
 src/sandboxes/runClaudeCodeAgent.ts | 15 ++++++++++++++-
 1 file changed, 14 insertions(+), 1 deletion(-)

diff --git a/src/sandboxes/runClaudeCodeAgent.ts b/src/sandboxes/runClaudeCodeAgent.ts
index 70a8a09..5714434 100644
--- a/src/sandboxes/runClaudeCodeAgent.ts
+++ b/src/sandboxes/runClaudeCodeAgent.ts
@@ -43,7 +43,20 @@ export async function runClaudeCodeAgent(
     },
   };
 
-  const command = await sandbox.runCommand(commandOpts as any);
+  let command;
+  try {
+    command = await sandbox.runCommand(commandOpts as any);
+  } catch (error) {
+    logStep(`${label} - sandbox.runCommand failed`, false, {
+      error: error instanceof Error ? error.message : String(error),
+      sandboxId: sandbox.sandboxId,
+      cwd,
+      cmd: "claude",
+      argsLength: args.length,
+      messageLength: message.length,
+    });
+    throw error;
+  }
   const result = await command.wait();
 
   const stdout = (await result.stdout()) || "";