Use sandbox client for builder validation

tests/test_builder_prompt.py

@@ -162,7 +162,7 @@ inline = Pricing(price_per_call_usd=0.01, compute={"runtime_seconds": 600}, tota
         self.assertIn("skills=skill_sources or None", files["deepagent-agent-design/SKILL.md"])
         self.assertIn("from a2a_pack import", files["a2apack-agent-authoring/SKILL.md"])
         self.assertIn("ctx.workspace_shell", files["a2apack-agent-authoring/SKILL.md"])
-        self.assertIn("create_deep_agent", files["deepagents-implementation-patterns/SKILL.md"])
+        self.assertIn("create_a2a_deep_agent", files["deepagents-implementation-patterns/SKILL.md"])
         self.assertIn('config={"recursion_limit": 500}', files["deepagents-implementation-patterns/SKILL.md"])
         self.assertIn("ctx.write_artifact", files["workspace-artifact-safety/SKILL.md"])
         self.assertIn("ctx.workspace_shell", files["workspace-artifact-safety/SKILL.md"])

tests/test_template_init.py

@@ -47,7 +47,7 @@ class TemplateInitTests(unittest.TestCase):
         self.assertIn("skills=skill_sources or None", files["agent.py"])
         self.assertIn("ctx.llm", files["agent.py"])
         self.assertIn("ctx.workspace_backend()", files["agent.py"])
-        self.assertIn("create_deep_agent", files["agent.py"])
+        self.assertIn("create_a2a_deep_agent", files["agent.py"])
         self.assertIn("name: research-agent", files["a2a.yaml"])
         self.assertIn("entrypoint: agent:ResearchAgent", files["a2a.yaml"])
         self.assertNotIn("{{ frontend_block }}", files["a2a.yaml"])
@@ -272,6 +272,35 @@ class TemplateInitTests(unittest.TestCase):
 
         self.assertIn("manifest_json must be a JSON object", result["error"])
 
+    def test_test_agent_in_sandbox_uses_attached_sandbox_client(self) -> None:
+        workspace = _FakeWorkspace()
+        prefix = "agents/demo-agent/"
+        workspace.write_bytes(prefix + "agent.py", b"print('ok')\n")
+        workspace.write_bytes(prefix + "a2a.yaml", b"name: demo-agent\n")
+        workspace.write_bytes(prefix + "requirements.txt", b"")
+        sandbox = _FakeSandbox()
+        tools = build_tools(
+            ToolContext(
+                bucket="bucket",
+                settings=_settings(),
+                workspace=workspace,
+                sandbox=sandbox,
+                grant_token="grant-token",
+            )
+        )
+        test_tool = _tool_by_name(tools, "test_agent_in_sandbox")
+
+        _FakeAsyncClient.posts = []
+        with patch("agent_builder.tools.httpx.AsyncClient", _FakeAsyncClient):
+            result = json.loads(
+                asyncio.run(test_tool.ainvoke({"name": "demo-agent"}))
+            )
+
+        self.assertEqual(result["exit_code"], 0)
+        self.assertEqual(len(sandbox.calls), 1)
+        self.assertEqual(sandbox.calls[0]["workspace"], "bucket")
+        self.assertEqual(_FakeAsyncClient.posts, [])
+
     def test_cp_deploy_tarball_posts_agent_dsl(self) -> None:
         workspace = _FakeWorkspace()
         prefix = "agents/demo-agent/"
@@ -529,6 +558,21 @@ class _FakeWorkspace:
         self.objects.pop(key, None)
 
 
+class _FakeExecResult:
+    exit_code = 0
+    stdout = "--- card ---\n{}"
+    stderr = ""
+
+
+class _FakeSandbox:
+    def __init__(self) -> None:
+        self.calls: list[dict[str, object]] = []
+
+    async def run_shell(self, script: str, **kwargs: object) -> _FakeExecResult:
+        self.calls.append({"script": script, **kwargs})
+        return _FakeExecResult()
+
+
 def _tool_by_name(tools: list[object], name: str) -> object:
     for tool in tools:
         if getattr(tool, "name", None) == name: