Require doc_type and summary in the AI output schema

scanfiler/ai/prompt.py

@@ -17,12 +17,14 @@
 - subdir: PREFER one of the existing subfolders below. Only set is_new_subdir=true and
   invent a new one when none genuinely fits.{new_rule}
 - date: the document's own date as ISO (YYYY, YYYY-MM, or YYYY-MM-DD); null if none.
-- doc_type: a short lowercase noun (receipt, invoice, medical_record, drawing, letter...).
-- summary: 1-2 sentences capturing what this document is, for later search.
+- doc_type: REQUIRED, never empty — a short lowercase noun (receipt, invoice,
+  medical_record, drawing, letter, statement...). Use "unknown" only as a last resort.
+- summary: REQUIRED, never empty — 1-2 sentences capturing what this document is, for
+  later search.
 - tags: 2-6 short lowercase keywords.
 - confidence: 0..1, your certainty in the filename+subdir. Be honest; low is fine.
 
-Existing subfolders: {subdirs}
+Every field is required. Existing subfolders: {subdirs}
 """
 
 _NO_NEW = "\n- New subfolders are DISABLED: you MUST pick from the existing list."

scanfiler/ai/schema.py

@@ -39,20 +39,26 @@ def build_response_format(existing_subdirs: list[str], allow_new: bool) -> dict:
     else:
         subdir_schema = {"type": "string"}
 
+    properties = {
+        "filename": {"type": "string", "minLength": 1},
+        "subdir": subdir_schema,
+        "is_new_subdir": {"type": "boolean"},
+        # minLength forces non-empty: the model can't satisfy the constraint by
+        # emitting "" for doc_type/summary (which it did when they were optional).
+        "doc_type": {"type": "string", "minLength": 1},
+        "date": {"type": ["string", "null"]},
+        "summary": {"type": "string", "minLength": 1},
+        "tags": {"type": "array", "items": {"type": "string"}},
+        "confidence": {"type": "number", "minimum": 0, "maximum": 1},
+    }
     schema = {
         "type": "object",
         "additionalProperties": False,
-        "required": ["filename", "subdir", "confidence"],
-        "properties": {
-            "filename": {"type": "string", "minLength": 1},
-            "subdir": subdir_schema,
-            "is_new_subdir": {"type": "boolean"},
-            "doc_type": {"type": "string"},
-            "date": {"type": ["string", "null"]},
-            "summary": {"type": "string"},
-            "tags": {"type": "array", "items": {"type": "string"}},
-            "confidence": {"type": "number", "minimum": 0, "maximum": 1},
-        },
+        # All fields required: forces doc_type/summary to be populated and makes the
+        # schema OpenAI strict-mode compliant (portable to cloud providers). `date`
+        # stays nullable; `tags` may be an empty array.
+        "required": list(properties),
+        "properties": properties,
     }
     return {
         "type": "json_schema",

tests/test_schema.py

@@ -24,3 +24,20 @@ def test_decision_defaults():
     assert d.is_new_subdir is False
     assert d.tags == []
     assert d.date is None
+
+
+def test_all_fields_required():
+    schema = build_response_format(["A"], allow_new=True)["json_schema"]["schema"]
+    # Every property is required (forces doc_type/summary; OpenAI strict-compliant).
+    assert set(schema["required"]) == set(schema["properties"])
+    assert "doc_type" in schema["required"]
+    assert "summary" in schema["required"]
+
+
+def test_doctype_and_summary_are_non_empty():
+    props = build_response_format([], allow_new=True)["json_schema"]["schema"]["properties"]
+    assert props["doc_type"]["minLength"] == 1
+    assert props["summary"]["minLength"] == 1
+    # date stays nullable, tags may be empty
+    assert props["date"] == {"type": ["string", "null"]}
+    assert "minItems" not in props["tags"]