Merge pull request #146 from JiusiServe/fix/knowledge-section-repeated-ids

sdk(knowledge): reject a section that heads one rule ID twice

Author: tzhouam

src/infermatrix_copilot/sdk/v1/knowledge.py

@@ -478,13 +478,24 @@ def build_prompt(self, batch: KnowledgeEvidenceBatch) -> str:
 
     @staticmethod
     def _section_rule_ids(section: str) -> tuple[str, ...]:
-        """The rule IDs every heading in a section carries, the declared
-        one first: a nested ``###`` heading is a rule heading too."""
-        ids: list[str] = []
-        for match in _RULE_HEADING_ID.finditer(section):
-            if match.group("rule") not in ids:
-                ids.append(match.group("rule"))
-        return tuple(ids)
+        """The rule IDs every heading in a section carries, in order, the
+        declared one first: a nested ``###`` heading is a rule heading too.
+        Repeats are kept so callers can reject a section that heads one ID
+        twice (``## X-2`` over ``### X-2``, or two ``### X-3``)."""
+        return tuple(
+            match.group("rule") for match in _RULE_HEADING_ID.finditer(section)
+        )
+
+    @classmethod
+    def _repeated_section_rule_id(cls, section: str) -> str:
+        """The first rule ID a section heads more than once, else ''."""
+        seen: set[str] = set()
+        for heading_id in cls._section_rule_ids(section):
+            if heading_id in seen:
+                return heading_id
+            seen.add(heading_id)
+        return ""
+
 
     @staticmethod
     def _rule_exists(page_text: str, rule_id: str) -> bool:
@@ -643,11 +654,18 @@ def validate_proposals(
                             "rule_id duplicates an earlier proposal on "
                             f"{proposed_ids[rule_id]}"
                         )
+                    if not reason and self._repeated_section_rule_id(section):
+                        reason = (
+                            "section heads rule_id "
+                            f"{self._repeated_section_rule_id(section)} more "
+                            "than once"
+                        )
                     if not reason:
                         # Every rule heading the section introduces, not
                         # only the declared one: a nested heading carrying
                         # another page's ID would otherwise land unchecked.
                         for heading_id in self._section_rule_ids(section):
+
                             owner = existing_ids.get(heading_id)
                             nested = heading_id != rule_id
                             if owner is None and nested and heading_id in proposed_ids:
@@ -948,6 +966,8 @@ def apply(
                     r"^##\s+", proposal.section_markdown, re.MULTILINE
                 )) != 1
                 or not 80 <= len(proposal.section_markdown) <= _MAX_SECTION_CHARS
+                or self._repeated_section_rule_id(proposal.section_markdown)
+
                 or not proposal.sources
                 or len(proposal.sources) > 10
                 or any(

test/test_sdk_knowledge_v1.py

@@ -788,3 +788,40 @@ def test_nested_headings_inside_a_section_are_checked_too(workspace):
     assert result.rejected[0].reason == (
         "nested rule heading X-2 already exists on knowledge/repos/x/rules.md"
     )
+
+
+@pytest.mark.parametrize(
+    "extra",
+    [
+        "\n\n### X-2 — the declared ID repeated in a nested heading\n\n- Required: reject. ^[PR #7]",
+        "\n\n### X-4 — first nested\n\n- Required: a. ^[PR #7]\n\n### X-4 — second nested\n\n- Required: b. ^[PR #7]",
+    ],
+    ids=["parent-and-nested", "two-nested"],
+)
+def test_section_heading_one_rule_id_twice_is_rejected(workspace, extra):
+    """A section may not head one ID twice, whether it repeats its own
+    declared ID under ``###`` or two nested headings share an ID: the
+    validators downstream do not enforce ID uniqueness."""
+    curator = KnowledgeCurator(workspace)
+    result = curator.validate_proposals(
+        _document(_rule(section=SECTION + extra)), _batch()
+    )
+
+    assert result.accepted == ()
+    repeated = "X-2" if "X-2 — the declared" in extra else "X-4"
+    assert result.rejected[0].reason == (
+        f"section heads rule_id {repeated} more than once"
+    )
+
+
+def test_apply_refuses_a_section_that_repeats_a_rule_id(workspace):
+    curator = KnowledgeCurator(workspace)
+    validation = curator.validate_proposals(_document(_rule()), _batch())
+    smuggled = replace(
+        validation.accepted[0],
+        section_markdown=SECTION + "\n\n### X-2 — repeated\n\n- Required: reject. ^[PR #7]",
+    )
+    tampered = replace(validation, accepted=(smuggled,))
+
+    with pytest.raises(KnowledgeCurationError, match="apply-time integrity"):
+        curator.apply(tampered, updated_on="2026-08-29")