first steps to better scanning of repos

tests/test_repository_scanner.py

@@ -42,6 +42,22 @@ def test_deterministic_scanner_extracts_structural_facts(tmp_path):
     assert languages == {"Python": 2}
 
 
+def test_scanner_records_scope_with_source_role(tmp_path):
+    repo = tmp_path / "sample"
+    repo.mkdir()
+    (repo / "SCOPE.md").write_text(
+        "# SCOPE\n\n## One-liner\n\nProvides OIDC profile enforcement.\n",
+        encoding="utf-8",
+    )
+
+    result = DeterministicScanner().scan(repo)
+
+    scope_fact = next(fact for fact in result.facts if fact.kind == "scope")
+    assert scope_fact.name == "SCOPE"
+    assert scope_fact.path == "SCOPE.md"
+    assert scope_fact.metadata["source_role"] == "scope_summary"
+
+
 def test_scanner_readme_only_fixture_records_docs_without_interfaces(tmp_path):
     repo = write_readme_only_repo(tmp_path)
 
@@ -116,3 +132,28 @@ def test_scanner_records_llm_provider_and_fallback_facts(tmp_path):
     assert ("credential_config", "Anthropic API key", ".env.example") in facts
     assert ("provider_registry", "LLM provider registry", "providers.py") in facts
     assert ("fallback_policy", "LLM provider fallback policy", "README.md") in facts
+
+
+def test_scanner_does_not_treat_agent_guidance_as_llm_provider(tmp_path):
+    repo = tmp_path / "key-cape-like"
+    repo.mkdir()
+    (repo / "README.md").write_text(
+        "# KeyCape\n\n"
+        "Backend adapters live in src/internal/adapters.\n\n"
+        "See `CLAUDE.md` for agent session protocol.\n",
+        encoding="utf-8",
+    )
+    (repo / "CLAUDE.md").write_text(
+        "# CLAUDE.md\n\n"
+        "This file provides guidance to Claude Code when working in this repo.\n",
+        encoding="utf-8",
+    )
+    (repo / "src").mkdir()
+    (repo / "src" / "go.mod").write_text("module keycape\n", encoding="utf-8")
+
+    result = DeterministicScanner().scan(repo)
+
+    facts = {(fact.kind, fact.name, fact.path) for fact in result.facts}
+    assert ("llm_provider", "Claude", "CLAUDE.md") not in facts
+    assert ("llm_provider", "Claude", "README.md") not in facts
+    assert ("provider_registry", "LLM provider registry", "README.md") not in facts