feat: add editable skill evaluation versions

2026-07-24 10:11:51 +08:00
parent ecc02c06b7
commit 6630e82069
5 changed files with 1070 additions and 3 deletions
@@ -169,6 +169,100 @@ class EvaluationRuntimeTests(unittest.TestCase):
            finally:
                runtime.shutdown()

+    def test_skill_versions_are_immutable_and_account_scoped(self) -> None:
+        with tempfile.TemporaryDirectory() as directory:
+            root = Path(directory)
+            _write_test_skill(root)
+            runtime = EvaluationRuntime(
+                root=root / 'evaluations',
+                cwd_for_account=lambda _account: root,
+                model_config_for=lambda _account: ModelConfig(model='test-model'),
+                account_paths_for=lambda _account: {'python_env': root / '.venv'},
+                max_workers=1,
+            )
+            try:
+                versions = runtime.list_skill_versions('alice', 'label-master')
+                self.assertEqual(len(versions['versions']), 1)
+                current_id = versions['current_snapshot_id']
+                self.assertTrue(versions['versions'][0]['is_current'])
+
+                manifest = runtime.get_skill_version_manifest(
+                    'alice',
+                    'label-master',
+                    current_id,
+                )
+                skill_file = next(
+                    item
+                    for item in manifest['files']
+                    if item['path'] == 'SKILL.md'
+                )
+                self.assertTrue(skill_file['editable'])
+                current_file = runtime.read_skill_version_file(
+                    'alice',
+                    'label-master',
+                    current_id,
+                    'SKILL.md',
+                )
+                updated_content = current_file['content'].replace(
+                    'Read the manifest',
+                    'Read the manifest carefully',
+                )
+                saved = runtime.save_skill_version(
+                    'alice',
+                    'label-master',
+                    base_snapshot_id=current_id,
+                    version_name='边界调整 v1',
+                    note='测试版本',
+                    files={'SKILL.md': updated_content},
+                )
+                self.assertEqual(saved['version_name'], '边界调整 v1')
+                self.assertEqual(saved['source_type'], 'local')
+                self.assertNotEqual(saved['id'], current_id)
+
+                original = runtime.read_skill_version_file(
+                    'alice',
+                    'label-master',
+                    current_id,
+                    'SKILL.md',
+                )
+                forked = runtime.read_skill_version_file(
+                    'alice',
+                    'label-master',
+                    saved['id'],
+                    'SKILL.md',
+                )
+                self.assertNotIn('carefully', original['content'])
+                self.assertIn('carefully', forked['content'])
+                dataset = runtime.create_dataset(
+                    account_id='alice',
+                    name='versioned',
+                    filename='versioned.csv',
+                    rows=[{'query': '导航去公司'}],
+                )
+                experiment = runtime.create_experiment(
+                    account_id='alice',
+                    dataset_id=dataset['id'],
+                    name='local version',
+                    snapshot_id=saved['id'],
+                )
+                self.assertEqual(experiment['snapshot_id'], saved['id'])
+                self.assertEqual(
+                    experiment['skill_version'],
+                    saved['content_hash'][:12],
+                )
+                with self.assertRaisesRegex(
+                    ValueError,
+                    'Skill 版本不存在',
+                ):
+                    runtime.read_skill_version_file(
+                        'bob',
+                        'label-master',
+                        saved['id'],
+                        'SKILL.md',
+                    )
+            finally:
+                runtime.shutdown()
+
    def test_result_normalization_and_label_equivalence(self) -> None:
        result = normalize_evaluation_output(
            json.dumps(