feat: add editable skill evaluation versions

tests/test_evaluation_api.py

@@ -178,6 +178,62 @@ class EvaluationApiTests(unittest.TestCase):
                 self.assertIn('text/csv', export.headers['content-type'])
                 self.assertIn('导航去公司', export.content.decode('utf-8-sig'))
 
+    def test_skill_version_endpoints(self) -> None:
+        with tempfile.TemporaryDirectory() as directory:
+            root = Path(directory)
+            state = _build_state(root)
+            with TestClient(create_app(state)) as client:
+                versions_response = client.get(
+                    '/api/evaluations/skills/label-master/versions',
+                    params={'account_id': 'alice'},
+                )
+                self.assertEqual(versions_response.status_code, 200)
+                versions = versions_response.json()
+                current_id = versions['current_snapshot_id']
+
+                manifest = client.get(
+                    (
+                        '/api/evaluations/skills/label-master/versions/'
+                        f'{current_id}'
+                    ),
+                    params={'account_id': 'alice'},
+                )
+                self.assertEqual(manifest.status_code, 200)
+                self.assertTrue(
+                    any(
+                        item['path'] == 'SKILL.md'
+                        for item in manifest.json()['files']
+                    )
+                )
+
+                file_response = client.get(
+                    (
+                        '/api/evaluations/skills/label-master/versions/'
+                        f'{current_id}/file'
+                    ),
+                    params={'account_id': 'alice', 'path': 'SKILL.md'},
+                )
+                self.assertEqual(file_response.status_code, 200)
+                content = file_response.json()['content'].replace(
+                    'Use the label catalog.',
+                    'Use the label catalog carefully.',
+                )
+                saved_response = client.post(
+                    '/api/evaluations/skills/label-master/versions',
+                    json={
+                        'account_id': 'alice',
+                        'base_snapshot_id': current_id,
+                        'version_name': '测试版本',
+                        'note': 'API 测试',
+                        'files': {'SKILL.md': content},
+                    },
+                )
+                self.assertEqual(saved_response.status_code, 200)
+                self.assertEqual(
+                    saved_response.json()['version_name'],
+                    '测试版本',
+                )
+
 
 if __name__ == '__main__':
     unittest.main()

tests/test_evaluation_runtime.py

@@ -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(