feat: add skill and assistant knowledge bases

offline-backend-20260512/backend/tests/test_skill_knowledge_scanner.py

@@ -0,0 +1,52 @@
+from pathlib import Path
+from zipfile import ZipFile
+
+from deerflow.skill_knowledge import build_skill_ir, scan_skill
+
+
+def test_scanner_excludes_code_blocks_source_and_sensitive_files(tmp_path: Path) -> None:
+    (tmp_path / "SKILL.md").write_text(
+        "# 示例\n本技能使用 客户画像数据。\n```python\nPASSWORD = 'must-not-enter-ir'\n```\n",
+        encoding="utf-8",
+    )
+    (tmp_path / "worker.py").write_text("# uses hidden system\nSECRET='never-read'", encoding="utf-8")
+    (tmp_path / ".env").write_text("TOKEN=super-secret-value", encoding="utf-8")
+
+    scan = scan_skill("example", tmp_path)
+    ir = build_skill_ir(scan)
+
+    assert scan.skipped_code_count == 1
+    assert scan.blocked_files == [{"path": ".env", "reason": "疑似密钥或敏感配置文件"}]
+    assert "must-not-enter-ir" not in str(ir)
+    assert "never-read" not in str(ir)
+    assert any(relation["predicate"] == "uses" for relation in ir["relations"])
+
+
+def test_code_only_change_does_not_change_digest(tmp_path: Path) -> None:
+    (tmp_path / "SKILL.md").write_text("# Stable knowledge", encoding="utf-8")
+    code = tmp_path / "worker.py"
+    code.write_text("print(1)", encoding="utf-8")
+    before = scan_skill("example", tmp_path).source_digest
+    code.write_text("print(2)", encoding="utf-8")
+    after = scan_skill("example", tmp_path).source_digest
+    assert before == after
+
+
+def test_zip_is_inspected_and_only_safe_knowledge_members_are_read(tmp_path: Path) -> None:
+    archive = tmp_path / "materials.zip"
+    with ZipFile(archive, "w") as output:
+        output.writestr("guide.md", "# 操作手册\n仅提炼这段知识。")
+        output.writestr("worker.py", "SECRET = 'code-must-not-be-read'")
+        output.writestr("../escape.md", "path traversal")
+
+    scan = scan_skill("example", tmp_path)
+
+    assert [file.path for file in scan.files] == ["materials.zip!/guide.md"]
+    assert scan.skipped_code_count == 1
+    assert scan.blocked_files == [
+        {
+            "path": "materials.zip!/../escape.md",
+            "reason": "压缩包包含路径穿越或符号链接",
+        }
+    ]
+    assert "code-must-not-be-read" not in str(scan.manifest)