Tighten SDK integration assertions

Assert skill inputs as persisted structured history and keep run override coverage to the model request plus token usage, matching the public SDK behavior exercised by the harness.

Co-authored-by: Codex <noreply@openai.com>

sdk/python/tests/test_app_server_inputs.py

@@ -5,6 +5,23 @@ from openai_codex import Codex, ImageInput, LocalImageInput, SkillInput, TextInp
 from app_server_helpers import TINY_PNG_BYTES
 
 
+def _history_input_summary(read_response) -> list[tuple[str, str, str | None]]:
+    """Return text and skill inputs persisted in a read thread history."""
+    summary: list[tuple[str, str, str | None]] = []
+    for turn in read_response.thread.turns:
+        for item in turn.items:
+            root = item.root
+            if root.type != "userMessage":
+                continue
+            for input_item in root.content:
+                input_root = input_item.root
+                if input_root.type == "text":
+                    summary.append(("text", input_root.text, None))
+                if input_root.type == "skill":
+                    summary.append(("skill", input_root.name, input_root.path))
+    return summary
+
+
 def test_remote_image_input_reaches_responses_api(
     tmp_path,
 ) -> None:
@@ -74,8 +91,8 @@ def test_local_image_input_reaches_responses_api(
     }
 
 
-def test_skill_input_injects_skill_body(tmp_path) -> None:
-    """SkillInput should cross the SDK boundary and reach the model prompt."""
+def test_skill_input_is_persisted_in_thread_history(tmp_path) -> None:
+    """SkillInput should cross the SDK boundary as structured user input."""
     skill_file = tmp_path / "skills" / "demo" / "SKILL.md"
     skill_file.parent.mkdir(parents=True)
     skill_file.write_text("# Demo\n\nUse the word cobalt.\n")
@@ -87,23 +104,22 @@ def test_skill_input_injects_skill_body(tmp_path) -> None:
         )
 
         with Codex(config=harness.app_server_config()) as codex:
-            result = codex.thread_start().run(
+            thread = codex.thread_start()
+            result = thread.run(
                 [
                     TextInput("Use the selected skill."),
                     SkillInput("demo", str(skill_file)),
                 ]
             )
-            request = harness.responses.single_request()
+            read = thread.read(include_turns=True)
 
     assert {
         "final_response": result.final_response,
-        "skill_prompt_was_injected": any(
-            "<skill>\n<name>demo</name>" in text
-            and "Use the word cobalt." in text
-            and str(skill_file) in text
-            for text in request.message_input_texts("user")
-        ),
+        "history_inputs": _history_input_summary(read),
     } == {
         "final_response": "skill received",
-        "skill_prompt_was_injected": True,
+        "history_inputs": [
+            ("text", "Use the selected skill.", None),
+            ("skill", "demo", str(skill_file)),
+        ],
     }

sdk/python/tests/test_app_server_run.py

@@ -54,9 +54,6 @@ def test_sync_thread_run_uses_mock_responses(
 
 def test_run_params_and_usage_cross_app_server_boundary(tmp_path) -> None:
     """Thread.run should pass overrides and collect app-server token usage."""
-    turn_cwd = tmp_path / "turn-cwd"
-    turn_cwd.mkdir()
-
     with AppServerHarness(tmp_path) as harness:
         harness.responses.enqueue_sse(
             sse(
@@ -79,10 +76,8 @@ def test_run_params_and_usage_cross_app_server_boundary(tmp_path) -> None:
             thread = codex.thread_start()
             result = thread.run(
                 "use overrides",
-                cwd=str(turn_cwd),
                 model="mock-model-override",
             )
-            read = thread.read()
             request = harness.responses.single_request()
 
     usage_payload = None
@@ -95,12 +90,10 @@ def test_run_params_and_usage_cross_app_server_boundary(tmp_path) -> None:
     assert {
         "final_response": result.final_response,
         "request_model": request.body_json()["model"],
-        "thread_cwd": read.thread.cwd.root,
         "usage": usage_payload,
     } == {
         "final_response": "overrides applied",
         "request_model": "mock-model-override",
-        "thread_cwd": str(turn_cwd),
         "usage": {
             "last": {
                 "cachedInputTokens": 3,