fix: require downloadable Work reports

tests/test_agent_loop.py

@@ -9,6 +9,7 @@ from agent_platform.auth import UserIdentity
 from agent_platform.models import get_model_spec
 from agent_platform.runtime.loop import (
     AgentLoop,
+    deliverable_evidence_failure,
     is_substantive_execution,
     non_source_execution_error,
     normalize_write_file_content,
@@ -68,6 +69,31 @@ def test_python_cannot_execute_documentation_or_data_files() -> None:
     assert non_source_execution_error({"command": "python3 -c 'print(1)' "}) is None
 
 
+def test_script_and_report_require_distinct_verified_artifacts() -> None:
+    assert (
+        deliverable_evidence_failure(
+            source_required=True,
+            report_required=True,
+            written_source_paths={"benchmark.py"},
+            written_report_paths=set(),
+            reports_written_after_execution=set(),
+            verified_paths={"benchmark.py"},
+        )
+        == "Create a separate report file such as `report.md`."
+    )
+    assert (
+        deliverable_evidence_failure(
+            source_required=True,
+            report_required=True,
+            written_source_paths={"benchmark.py"},
+            written_report_paths={"report.md"},
+            reports_written_after_execution={"report.md"},
+            verified_paths={"benchmark.py", "report.md"},
+        )
+        is None
+    )
+
+
 def test_read_only_shell_commands_do_not_satisfy_execution_evidence() -> None:
     assert not is_substantive_execution("exec", {"command": "cat report.txt"})
     assert not is_substantive_execution("exec", {"command": "sed -n '1,20p' script.py"})
@@ -355,6 +381,106 @@ async def test_artifact_completion_is_rejected_until_written_and_verified(settin
         await store.close()
 
 
+async def test_script_and_report_completion_requires_both_files(settings) -> None:
+    responses = [
+        {
+            "choices": [
+                {
+                    "message": {
+                        "role": "assistant",
+                        "tool_calls": [
+                            tool_call("1", "write_file", {"path": "benchmark.py", "content": "print('1 ms')\n"})
+                        ],
+                    },
+                    "finish_reason": "tool_calls",
+                }
+            ]
+        },
+        {
+            "choices": [
+                {
+                    "message": {
+                        "role": "assistant",
+                        "tool_calls": [tool_call("2", "read_file", {"path": "benchmark.py"})],
+                    },
+                    "finish_reason": "tool_calls",
+                }
+            ]
+        },
+        {
+            "choices": [
+                {
+                    "message": {
+                        "role": "assistant",
+                        "tool_calls": [tool_call("3", "exec", {"command": "python3 benchmark.py"})],
+                    },
+                    "finish_reason": "tool_calls",
+                }
+            ]
+        },
+        {"choices": [{"message": {"role": "assistant", "content": "完成"}, "finish_reason": "stop"}]},
+        {
+            "choices": [
+                {
+                    "message": {
+                        "role": "assistant",
+                        "tool_calls": [
+                            tool_call("4", "write_file", {"path": "report.md", "content": "# Report\n\n1 ms\n"})
+                        ],
+                    },
+                    "finish_reason": "tool_calls",
+                }
+            ]
+        },
+        {
+            "choices": [
+                {
+                    "message": {
+                        "role": "assistant",
+                        "tool_calls": [tool_call("5", "read_file", {"path": "report.md"})],
+                    },
+                    "finish_reason": "tool_calls",
+                }
+            ]
+        },
+        {"choices": [{"message": {"role": "assistant", "content": "已完成两个文件"}, "finish_reason": "stop"}]},
+    ]
+    registry = RecordingRegistry()
+    store = RuntimeStore(settings.database_url)
+    await store.initialize()
+    events = []
+
+    async def callback(event_type, payload):
+        events.append(event_type)
+
+    try:
+        answer = await AgentLoop(
+            ScriptedProvider(responses),
+            registry,
+            store,
+            max_tool_output_chars=10_000,
+        ).run(
+            spec=get_model_spec("work-light"),
+            messages=[{"role": "user", "content": "写脚本对比算法并给一个报告"}],
+            identity=UserIdentity("u1", "", "", "user"),
+            raw_user_jwt="jwt",
+            chat_id="script-report",
+            callback=callback,
+        )
+    finally:
+        await store.close()
+
+    assert answer == "已完成两个文件"
+    assert events.count("completion.rejected") == 1
+    assert [name for name, _ in registry.calls] == [
+        "write_file",
+        "read_file",
+        "exec",
+        "write_file",
+        "read_file",
+    ]
+
+
 class FailingExecutionRegistry(RecordingRegistry):
     def __init__(self) -> None:
         super().__init__()
@@ -441,7 +567,7 @@ async def test_failed_execution_must_be_repaired_before_completion(settings) ->
     try:
         answer = await AgentLoop(provider, registry, store, max_tool_output_chars=10_000).run(
             spec=get_model_spec("work-light"),
-            messages=[{"role": "user", "content": "写脚本对比排序算法并生成报告"}],
+            messages=[{"role": "user", "content": "写脚本对比排序算法"}],
             identity=UserIdentity("u1", "", "", "user"),
             raw_user_jwt="jwt",
             chat_id="c1",