ai-agent-book 精选快照（<2MB 代码与文档，来自 github.com/bojieli/ai-agent-book）

chapter4/perception-tools/test_experiment_4_1.py

@@ -0,0 +1,147 @@
+"""Offline acceptance checks for the exact Experiment 4-1 campaign."""
+
+from __future__ import annotations
+
+import importlib.util
+import json
+import sys
+from pathlib import Path
+
+HERE = Path(__file__).resolve().parent
+RUNNER = HERE / "run_experiment_4_1.py"
+SPEC = importlib.util.spec_from_file_location("experiment_4_1_runner", RUNNER)
+runner = importlib.util.module_from_spec(SPEC)
+assert SPEC.loader is not None
+sys.modules[SPEC.name] = runner
+SPEC.loader.exec_module(runner)
+
+
+def _protocol() -> dict:
+    return json.loads((HERE / "experiment_protocol.json").read_text(encoding="utf-8"))
+
+
+def _receipt(case: str, *, success: bool = True) -> dict:
+    return {
+        "case": case,
+        "tool": runner.CASE_TO_TOOL[case],
+        "transport": "mcp-stdio",
+        "mcp_result_is_error": False,
+        "success": success,
+        "substantive_observation": success,
+        "backend_provenance": runner.PROVENANCE[runner.CASE_TO_TOOL[case]],
+        "simulation_markers": [],
+        "error_type": None,
+        "payload": {"success": success},
+    }
+
+
+def _all_receipts() -> list[dict]:
+    protocol = _protocol()
+    return [
+        _receipt(case)
+        for category in protocol["categories"].values()
+        for case in category.get("required_cases", []) + category.get("required_safety_cases", [])
+    ]
+
+
+def _catalog() -> dict:
+    names = set(runner.CASE_TO_TOOL.values())
+    names.update(f"extra_{index}" for index in range(120))
+    return {
+        "transport": "mcp-stdio",
+        "tools_list_received": True,
+        "mcp_sdk_version": "2.0.0",
+        "protocol_version": "2026-07-28",
+        "tool_count": len(names),
+        "unique_tool_count": len(names),
+        "tool_names": sorted(names),
+    }
+
+
+def test_catalog_gate_requires_v2_sdk_and_current_protocol():
+    catalog = _catalog()
+    assert runner.derive_acceptance(
+        _protocol(), catalog, _all_receipts(), outside_witness_unchanged=True
+    )["gates"]["catalog_from_real_mcp"]
+
+    catalog["protocol_version"] = "2025-11-25"
+    assert not runner.derive_acceptance(
+        _protocol(), catalog, _all_receipts(), outside_witness_unchanged=True
+    )["gates"]["catalog_from_real_mcp"]
+
+    catalog.update(protocol_version="2026-07-28", mcp_sdk_version="1.29.0")
+    assert not runner.derive_acceptance(
+        _protocol(), catalog, _all_receipts(), outside_witness_unchanged=True
+    )["gates"]["catalog_from_real_mcp"]
+
+
+def test_protocol_covers_every_manuscript_category_and_mutation():
+    protocol = _protocol()
+    assert list(protocol["categories"]) == [
+        "search", "multimodal", "filesystem", "public_data", "private_data"
+    ]
+    assert {"filesystem_move", "filesystem_copy", "filesystem_delete"} <= set(
+        protocol["categories"]["filesystem"]["required_cases"]
+    )
+    assert {"calendar_events", "notion_search"} == set(
+        protocol["categories"]["private_data"]["required_cases"]
+    )
+
+
+def test_acceptance_fails_closed_when_receipts_are_missing():
+    result = runner.derive_acceptance(
+        _protocol(), _catalog(), [], outside_witness_unchanged=True
+    )
+    assert result["status"] == "failed"
+    assert not result["gates"]["exact_case_set_recorded"]
+    assert not result["gates"]["private_data_category_passed"]
+
+
+def test_private_credential_failure_is_blocked_and_never_passed():
+    receipts = _all_receipts()
+    for receipt in receipts:
+        if receipt["case"] in {"calendar_events", "notion_search"}:
+            receipt.update({
+                "success": False,
+                "substantive_observation": False,
+                "error_type": "missing_credentials",
+                "payload": {
+                    "success": False,
+                    "metadata": {"error_type": "missing_credentials"},
+                },
+            })
+        if receipt["case"].startswith("reject_"):
+            receipt.update({
+                "success": False,
+                "substantive_observation": False,
+                "error_type": "PermissionError",
+                "payload": {"success": False},
+            })
+    result = runner.derive_acceptance(
+        _protocol(), _catalog(), receipts, outside_witness_unchanged=True
+    )
+    assert result["status"] == "blocked"
+    assert result["categories"]["private_data"]["status"] == "blocked"
+    assert result["gates"]["private_data_category_passed"] is False
+
+
+def test_mock_marker_invalidates_an_apparent_success():
+    receipt = _receipt("weather")
+    receipt["simulation_markers"] = ["mock"]
+    assert runner.valid_success(receipt) is False
+
+
+def test_isolation_probe_must_preserve_outside_witness():
+    receipts = _all_receipts()
+    for receipt in receipts:
+        if receipt["case"].startswith("reject_"):
+            receipt.update({
+                "success": False,
+                "substantive_observation": False,
+                "error_type": "PermissionError",
+            })
+    result = runner.derive_acceptance(
+        _protocol(), _catalog(), receipts, outside_witness_unchanged=False
+    )
+    assert result["status"] == "failed"
+    assert result["gates"]["filesystem_isolation_probes_rejected"] is False