improved scanner

src/repo_registry/candidate_graph/generator.py

@@ -103,7 +103,10 @@ class CandidateGraphGenerator:
             capabilities.append(
                 self._interface_capability(interfaces, tests, examples, docs, chunks)
             )
-        if llm_providers or provider_registries or fallback_policies:
+        promotable_llm_facts = self._promotable_llm_facts(
+            llm_providers + provider_registries + fallback_policies
+        )
+        if promotable_llm_facts:
             capabilities.append(
                 self._llm_provider_capability(
                     llm_providers,
@@ -269,6 +272,8 @@ class CandidateGraphGenerator:
                 credentials,
                 registries,
                 fallback_policies,
+            ) + self._utility_relationship_attributes(
+                providers + credentials + registries + fallback_policies
             ),
             features=features,
             evidence=self._evidence(tests, examples, docs),
@@ -761,6 +766,37 @@ class CandidateGraphGenerator:
     def _facts(self, facts: list[ObservedFact], kind: str) -> list[ObservedFact]:
         return [fact for fact in facts if fact.kind == kind]
 
+    def _promotable_llm_facts(self, facts: list[ObservedFact]) -> list[ObservedFact]:
+        return [
+            fact
+            for fact in facts
+            if self._utility_relationship(fact) in {"owned", "facade", "adapter"}
+        ]
+
+    def _utility_relationship(self, fact: ObservedFact) -> str:
+        relationship = fact.metadata.get("utility_relationship")
+        if isinstance(relationship, str) and relationship:
+            return relationship
+        source_role = fact.metadata.get("source_role")
+        if source_role == "implementation_source":
+            lower_path = fact.path.lower()
+            if "adapter" in lower_path or "provider" in lower_path:
+                return "adapter"
+            return "owned"
+        if source_role == "configuration":
+            return "configure"
+        if source_role == "dependency_declaration":
+            return "dependency"
+        if source_role in {"agent_guidance", "ci_tooling"}:
+            return "tooling"
+        if not source_role and fact.path.lower().endswith((".py", ".ts", ".js")):
+            return "owned"
+        return "mention"
+
+    def _utility_relationship_attributes(self, facts: list[ObservedFact]) -> list[str]:
+        relationships = sorted({self._utility_relationship(fact) for fact in facts})
+        return [f"utility-{relationship}" for relationship in relationships]
+
     def _source_refs(self, facts: list[ObservedFact]) -> list[SourceReference]:
         return [
             SourceReference(

src/repo_registry/repo_scanning/scanner.py

@@ -358,6 +358,10 @@ class DeterministicScanner:
             source_role = self._source_role(relative)
             if source_role == "agent_guidance":
                 continue
+            utility_relationship = self._provider_utility_relationship(
+                source_role,
+                relative,
+            )
             for needle, provider in LLM_PROVIDER_HINTS.items():
                 if not self._has_provider_signal(lower_text, needle):
                     continue
@@ -372,6 +376,7 @@ class DeterministicScanner:
                         metadata={
                             "source": "provider_hint",
                             "source_role": source_role,
+                            "utility_relationship": utility_relationship,
                         },
                     ),
                 )
@@ -389,6 +394,7 @@ class DeterministicScanner:
                         metadata={
                             "source": "environment_variable",
                             "source_role": source_role,
+                            "utility_relationship": "configure",
                         },
                     ),
                 )
@@ -412,6 +418,7 @@ class DeterministicScanner:
                             metadata={
                                 "source": "provider_registry_hint",
                                 "source_role": source_role,
+                                "utility_relationship": utility_relationship,
                             },
                         ),
                     )
@@ -429,11 +436,30 @@ class DeterministicScanner:
                         metadata={
                             "source": "fallback_hint",
                             "source_role": source_role,
+                            "utility_relationship": utility_relationship,
                         },
                     ),
                 )
         return facts
 
+    def _provider_utility_relationship(
+        self,
+        source_role: str,
+        relative_path: str,
+    ) -> str:
+        if source_role == "implementation_source":
+            lower = relative_path.lower()
+            if "adapter" in lower or "provider" in lower:
+                return "adapter"
+            return "owned"
+        if source_role == "configuration":
+            return "configure"
+        if source_role == "dependency_declaration":
+            return "dependency"
+        if source_role in {"ci_tooling", "agent_guidance"}:
+            return "tooling"
+        return "mention"
+
     def _source_role(self, relative_path: str) -> str:
         lower = relative_path.lower()
         parts = lower.split("/")