fix: harden task2 preprocessing and vector guards

harness/tests/test_qdrant_cli_commands.py

@@ -225,7 +225,7 @@ def test_vector_index_schema_json_failure_is_safe(monkeypatch, tmp_path):
     _write_schema_artifacts(tmp_path)
     monkeypatch.setattr(
         "tht.adapters.factory.build_vector_store",
-        lambda cfg, require_write: (_ for _ in ()).throw(RuntimeError("secret qdrant endpoint")),
+        lambda cfg, require_write: (_ for _ in ()).throw(Exception("secret qdrant endpoint")),
     )
     response = CliRunner().invoke(app, ["vector", "index-schema", "--json", "-c", str(cfg)])
     assert response.exit_code != 0
@@ -233,6 +233,7 @@ def test_vector_index_schema_json_failure_is_safe(monkeypatch, tmp_path):
     payload = json.loads(response.stdout)
     assert payload == {"status": "failed", "code": "schema_index_failed"}
     assert "secret qdrant" not in response.stdout
+    assert response.stderr == ""
 
 
 
@@ -294,3 +295,39 @@ def test_vector_index_schema_json_legacy_config_has_no_stderr_on_failure(tmp_pat
     assert json.loads(response.stdout) == {
         "status": "failed", "code": "physical_schema_missing"
     }
+
+
+def test_vector_index_schema_guards_before_artifact_access(tmp_path, monkeypatch):
+    import tht.cli.vector_cmd as command
+
+    cfg = _legacy_qdrant_runtime_config(tmp_path)
+    text = cfg.read_text()
+    text = text.replace("vectors:\n  type: qdrant\n  base_url: http://qdrant:6333\n  collection: psd-clinical\n", "")
+    text = text.replace("embeddings:\n  provider: ollama_internal\n  base_url: http://embedding:11434\n  model: qwen3-embedding:0.6b\n  dim: 1024\n", "")
+    cfg.write_text(text)
+    monkeypatch.setattr(command, "_load_schema_artifacts", lambda cfg: (_ for _ in ()).throw(AssertionError("artifact access")))
+    response = CliRunner().invoke(app, ["vector", "index-schema", "--json", "-c", str(cfg)])
+
+    assert response.exit_code == 1
+    assert json.loads(response.stdout) == {"status": "failed", "code": "vector_configuration_missing"}
+    assert response.stderr == ""
+
+
+def test_vector_index_schema_core_reuses_injected_artifacts_without_path_resolution(tmp_path, monkeypatch):
+    import tht.cli.vector_cmd as command
+    from tht.config import load_config
+    from tht.mschema.models import Annotations, PhysicalSchema
+
+    cfg_path = _qdrant_runtime_config(tmp_path)
+    _write_schema_artifacts(tmp_path)
+    physical = PhysicalSchema.from_yaml(tmp_path / "artifacts" / "mschema" / "physical.yaml")
+    annotations = Annotations.from_yaml(tmp_path / "artifacts" / "mschema" / "annotations.yaml")
+    store = _FakeVectorStore()
+    monkeypatch.setattr(command, "physical_path", lambda cfg: (_ for _ in ()).throw(AssertionError("physical path")))
+    monkeypatch.setattr(command, "annotations_path", lambda cfg: (_ for _ in ()).throw(AssertionError("annotations path")))
+    monkeypatch.setattr("tht.adapters.factory.build_vector_store", lambda cfg, require_write: store)
+    monkeypatch.setattr(command, "make_embedder", lambda _: _FakeEmbedder())
+
+    payload = command.index_schema_data(load_config(cfg_path), physical=physical, annotations=annotations)
+
+    assert payload["status"] == "succeeded"