@zhuoyuezs/ml-platform 0.1.5 → 0.1.7

package/DEVELOPMENT.md

@@ -14,7 +14,8 @@
   用户级 Skill，不要求额外安装参数。
 - 管理命令默认输出简洁结果；Agent 和 CI 显式使用 `--json` 获取稳定结构化输出。
 - npm `postinstall` 不得修改 Agent Skills 目录。
-- Python business client 只保留为迁移 oracle，不进入 release。
+- Python business client 与迁移 oracle 已删除；release builder 继续防御性排除 Python
+  runtime、wheel 和旧 wrapper，npm 包只携带 JavaScript CLI 与纯 Skill payload。
 - Codex CLI 不重新打包；当前只支持唯一自研 Host。
 - builder 使用 Python，因此开发构建命令仍通过 `uv run ...` 启动；最终 npm 包和用户侧
   `ml-platform` 运行时只依赖 Node.js，不依赖 Python 或 `uv`。
@@ -48,7 +49,7 @@ package/
 └── DEVELOPMENT.md
 ```
 
-严格 `files` allowlist 必须排除 `node_modules`、测试 oracle、旧 wheel、Skill runtime、旧 wrapper、
+严格 `files` allowlist 必须排除 `node_modules`、旧 wheel、Skill runtime、旧 wrapper、
 Host 副本、日志、会话数据、attic、凭证和实验文件。
 
 最终发布单元是 builder 生成并通过 smoke test 的 `.tgz`，不是 `dist/package/` 目录。目录只用于
@@ -61,24 +62,25 @@ Host 副本、日志、会话数据、attic、凭证和实验文件。
 
 ### Step 1：完成 business-client-js 迁移（已完成）
 
-- 补齐全部 Python CLI 命令和参数契约；
+- 完成原 Python CLI 命令和参数契约到 JavaScript 的迁移；
 - 对齐 JSON stdout、stderr 和退出码；
 - 补齐数值参数、子命令 help、路径展开和 HTTP 错误测试；
 - 已注册 Dataset 支持 `get-dataset`、`resolve-dataset` 和
   `build-registered-dataset` API-first 流程，不依赖本地 Manifest；
 - Catalog 规范化保留 `project`，Registry 只读元数据与可发布契约分离，目标 project
   不一致时在访问 Registry 前失败；
-- 扩大 Catalog normalization differential tests；
+- 扩大 Catalog normalization 与 OpenAPI contract tests；
 - 完成 Job、Artifact、inference 和 Operator upload 负向测试；
 - 删除正式运行路径中的旧产品名兼容；
 - 独立执行 `npm pack --json --dry-run` 机器可读清单校验、clean install 和 CLI smoke test。
 
-完成标准：Python 只剩 oracle 作用，正式 CLI 不需要 Python、`uv` 或第三方 npm runtime 依赖。
+完成标准：正式 CLI 不需要 Python、`uv` 或第三方 npm runtime 依赖，旧 Python client 可删除。
 
 ### Step 2：重建纯 feature-management Skill（已完成）
 
 - 保留 `SKILL.md`、`agents/`、`references/` 和 `assets/catalog-template/`；
-- 删除 `runtime/business-client-js/`、`assets/wheels/`、Skill installer 和 CLI launcher；
+- npm staging 时排除 `assets/wheels/`、Skill installer 和 CLI launcher；这些旧 Python
+  runtime/installer/zip 交付物随后已从仓库彻底删除，staging 排除逻辑保留为防御性边界。
 - 所有命令示例改为直接调用 `ml-platform`；
 - 明确 CLI 是外部 prerequisite，并提供 `ml-platform --help` 自省路径；
 - 更新 `agents/openai.yaml`；
@@ -134,7 +136,7 @@ Host 副本、日志、会话数据、attic、凭证和实验文件。
 
 - `git diff --check -- npm business-client-js skills/feature-management agent_poc/codex-host`；
 - JS business-client tests；
-- Python oracle tests；
+- JavaScript OpenAPI contract 与 Catalog tests；
 - Skill validator 和 forward tests；
 - npm installer tests；
 - release builder；

package/README.md

@@ -120,7 +120,7 @@ npx --yes --registry=https://registry.npmjs.org/ \
 ## 包内容
 
 - 一份 JavaScript business CLI；
-- 一份纯 `feature-management` Skill；
+- 两份纯 Skill：`feature-management` 与 `model-lifecycle-management`；
 - 显式安装、升级、状态、校验和回滚逻辑。
 
 不包含 Codex CLI、Python business-client wheel、Agent Host、平台服务端、模型训练或 Kubernetes

package/checksums.json

@@ -2,8 +2,8 @@
   "files": [
     {
       "path": "runtime/business-client/README.md",
-      "sha256": "sha256:72b0cbbb18a7d0f1bca5d97706588fbebcd2781833ae023bae0317be13342601",
-      "size_bytes": 1880
+      "sha256": "sha256:b12f30609c756ac4473723df434e6b1055f8713b7acfc3278b091b52288f6347",
+      "size_bytes": 2459
     },
     {
       "path": "runtime/business-client/package-lock.json",
@@ -12,18 +12,18 @@
     },
     {
       "path": "runtime/business-client/package.json",
-      "sha256": "sha256:b784fa27db6e40f72da11b9bca007155146174bc33edc4511c61d4ea4af93cf1",
-      "size_bytes": 508
+      "sha256": "sha256:9e815ffe0487eb216447f4a4b31d248fe37392e08f0fd4bae2aba9c4f4285547",
+      "size_bytes": 501
     },
     {
       "path": "runtime/business-client/src/catalog.js",
-      "sha256": "sha256:afc2ba28eee2d1dcd833a2e5c4ac923421dc60302b31bf985aa638849bbf3d36",
-      "size_bytes": 54246
+      "sha256": "sha256:8dcc4c4baff7f292a6fb09d6f82693b5ce7da16b9f0a9e03b22e1f18e893240f",
+      "size_bytes": 53659
     },
     {
       "path": "runtime/business-client/src/cli.js",
-      "sha256": "sha256:f2634e346a4c5bdc42c81e316791906f809b768a77672e36d29cc51410c010ed",
-      "size_bytes": 29818
+      "sha256": "sha256:ca7bbee61e83f4597c0da75fd3527f1ee43d5257fff07693b581987efeeaafd3",
+      "size_bytes": 41926
     },
     {
       "path": "runtime/business-client/src/config.js",
@@ -37,8 +37,8 @@
     },
     {
       "path": "skills/feature-management/SKILL.md",
-      "sha256": "sha256:7b58cc4ff2a7b83a29a6dbf9ad54e9e4a02efd332cd01c5d8fb83020b51ea506",
-      "size_bytes": 29676
+      "sha256": "sha256:ba725b6be2d0f8e4d741d3a6f96e64657112c15935f8e9fd2818cd0b17380816",
+      "size_bytes": 30428
     },
     {
       "path": "skills/feature-management/agents/openai.yaml",
@@ -92,8 +92,8 @@
     },
     {
       "path": "skills/feature-management/references/commands.md",
-      "sha256": "sha256:c20ccf9ad8d505a1880a7e35773edb5040bbbcfeba3a64316e3148a6e337bdc2",
-      "size_bytes": 11546
+      "sha256": "sha256:c748f1877f6d2ac1f6ecc29e45cd270997cbd35b498a190decddb66c74b840ab",
+      "size_bytes": 13751
     },
     {
       "path": "skills/feature-management/references/contracts.md",
@@ -109,6 +109,31 @@
       "path": "skills/feature-management/references/platform-capability-guide.md",
       "sha256": "sha256:40d0f43a30688c5852ef09203ec35b4dc2c6b9db4bc7dd26dfae52e12d10e9ee",
       "size_bytes": 4257
+    },
+    {
+      "path": "skills/model-lifecycle-management/SKILL.md",
+      "sha256": "sha256:48c60d20ab9203cde9681fc61a0515ce3f316ad674e4f349aa21e320127a7520",
+      "size_bytes": 2837
+    },
+    {
+      "path": "skills/model-lifecycle-management/agents/openai.yaml",
+      "sha256": "sha256:37857849a8baa6b06b6d5d07ae6036080239356fcccd36f77bbdc52c1dfa283f",
+      "size_bytes": 249
+    },
+    {
+      "path": "skills/model-lifecycle-management/references/evaluation.md",
+      "sha256": "sha256:0139e5fc79beeab84fb31792bdac37c3331676e1f591c57390ecc6336873dcc9",
+      "size_bytes": 1797
+    },
+    {
+      "path": "skills/model-lifecycle-management/references/packaging.md",
+      "sha256": "sha256:c987d4095f23a1f18ba55effb14f946e0f4cdcb661d7af28af560cc8f72d8b2c",
+      "size_bytes": 1101
+    },
+    {
+      "path": "skills/model-lifecycle-management/references/training.md",
+      "sha256": "sha256:e6971f2ef9c2b8645388dd4bac4562408f7a7b1adb0b7171c7e6b4912ea48b50",
+      "size_bytes": 1852
     }
   ],
   "schema_version": "data_platform.ml_platform_checksums/v1"

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@zhuoyuezs/ml-platform",
-  "version": "0.1.5",
+  "version": "0.1.7",
   "description": "ML 数据平台 business CLI 与 Agent Skills 管理工具",
   "bin": {
     "ml-platform": "scripts/main.js"

package/release.json

@@ -13,7 +13,7 @@
     "entrypoint": "src/cli.js",
     "name": "ml-platform",
     "path": "runtime/business-client",
-    "sha256": "sha256:b509f9ccdedc4aa5d4d01c77609e4a16020ac6f74c62efc3af861877d897fe94",
+    "sha256": "sha256:bbb3c59d5d7e16f92e6496e5596e59af3af4fdf4f820487d9ceb6eb85ca0857b",
     "version": "0.7.0"
   },
   "policy_sha256": "sha256:02fbde0134696b78c0c365d6e0b76c9814227208da5b2b3d4e5dbd82dc4a3606",
@@ -25,7 +25,7 @@
     "ml_data_platform.feature_set/v1",
     "ml_data_platform.dataset_manifest/v1"
   ],
-  "release_version": "0.1.5",
+  "release_version": "0.1.7",
   "runtime_requirements": {
     "node": ">=18",
     "os": [
@@ -38,9 +38,15 @@
     "feature-management": {
       "path": "skills/feature-management",
       "requires_cli": ">=0.7.0 <0.8.0",
-      "revision": "0.1.5",
-      "sha256": "sha256:ea7ef622f10300a3b5ee200adcf10070d18fadfb899d1bf2cc031ef42ffc5c03"
+      "revision": "0.1.7",
+      "sha256": "sha256:15131c1303b143373e4b8671301f5b884013a1906a44cfc85dd6f93f43c5d8c8"
+    },
+    "model-lifecycle-management": {
+      "path": "skills/model-lifecycle-management",
+      "requires_cli": ">=0.7.0 <0.8.0",
+      "revision": "0.1.7",
+      "sha256": "sha256:b6eac3ac75a82cf27db2efe2daf2b5b8777c95a0841d48f93ada29ef08b8d8fe"
     }
   },
-  "source_commit": "d18a5b535a2a6c41fe4732e8e73739f2728c1e58"
+  "source_commit": "62db427abfd17b9c8dc58964b03aec21927217b5"
 }

package/runtime/business-client/README.md

@@ -1,6 +1,7 @@
 # Business Client JS
 
-`business-client/` 的 JavaScript 迁移实现。Python client 目前仅作为迁移期 parity oracle；
+工业时序智能建模平台唯一受支持的业务 CLI 与 API client 实现。客户端合同由平台
+OpenAPI 快照、JavaScript 单元测试和 npm 发布测试直接验证；
 本目录是业务 CLI 的唯一 JavaScript 实现源码，Skill 通过外部 `ml-platform` 命令调用它，
 不复制或内嵌本目录的 runtime。
 
@@ -35,6 +36,17 @@ ml-platform list-feature-versions <name> --project <project> --all
 ml-platform list-jobs --project <project> --status succeeded --all
 ```
 
+平台资源生命周期与部署绑定的因果上下文读取也由同一 CLI 覆盖：
+
+```bash
+ml-platform list-parameter-references <name> <version> --project <project>
+ml-platform get-parameter-lifecycle <name> <version> --project <project>
+ml-platform fetch-inference-context /absolute/path/to/request.json
+```
+
+Parameter、Feature、FeatureSet 的删除命令执行软删除，并支持带 actor、reason 和幂等键的
+恢复；调用方应在任何变更前查询直接与传递引用。
+
 仓库中的 `api-spec/latest.json` 从 `origin/dev` 生成。同步或验证契约时运行：
 
 ```bash

package/runtime/business-client/package.json

@@ -2,7 +2,7 @@
   "name": "@zhuoyuezs/ml-platform-business-client",
   "version": "0.7.0",
   "private": true,
-  "description": "JavaScript parity implementation of the ML Platform business client",
+  "description": "JavaScript implementation of the ML Platform business client",
   "bin": {
     "ml-platform": "src/cli.js"
   },

package/runtime/business-client/src/catalog.js

@@ -58,7 +58,6 @@ function normalizeQualityRules(value) { exactKeys(value, ["valid_range", "allow_
 function normalizeCodeArtifact(value) { object(value, "operator code_artifact"); const sha256 = requiredString(value.sha256, "code_artifact sha256"); if (!wheelDigestPattern.test(sha256)) throw new Error("code_artifact sha256 is invalid"); const filename = requiredString(value.filename, "code_artifact filename"); if (path.basename(filename) !== filename || !wheelFilenamePattern.test(filename)) throw new Error("code_artifact filename is invalid"); if (!Number.isInteger(value.size_bytes) || value.size_bytes <= 0) throw new Error("code_artifact size_bytes must be a positive integer"); if (defaultIfAbsent(value, "format", "wheel") !== "wheel") throw new Error("code_artifact format must be wheel"); return { uri: requiredString(value.uri, "code_artifact uri"), sha256, format: "wheel", filename, size_bytes: value.size_bytes, python_version: optionalString(value.python_version, "code_artifact python_version"), dependency_lock_hash: optionalString(value.dependency_lock_hash, "code_artifact dependency_lock_hash") }; }
 
 function normalizeParameter(payload) {
-  exactKeys(payload, ["schema_version", "project", "name", "display_name", "version", "data_type", "unit", "expected_frequency", "source", "time_semantics", "availability_sla", "value_field", "quality_rules", "rounding", "owner"], ["name", "display_name", "version", "source"], "parameter");
   const source = object(payload.source, "parameter source");
   if (Object.prototype.hasOwnProperty.call(source, "parameters")) throw new Error("parameter source uses unsupported field 'parameters'; use 'params'");
   if (source.mode === "sql" && typeof source.sql === "string" && source.sql.trimEnd().endsWith(";")) throw new Error("SQL Parameter source.sql must not end with a semicolon");
@@ -66,7 +65,7 @@ function normalizeParameter(payload) {
   return { schema_version: pythonString(defaultIfAbsent(payload, "schema_version", "ml_data_platform.parameter/v1")), project: identifier(defaultIfAbsent(payload, "project", "default"), "parameter project"), name: identifier(payload.name, "parameter name"), display_name: requiredString(payload.display_name, "parameter display_name"), version: identifier(payload.version, "parameter version"), data_type: pythonString(defaultIfAbsent(payload, "data_type", "time_series")), unit: pythonString(defaultIfAbsent(payload, "unit", "")), expected_frequency: pythonString(defaultIfAbsent(payload, "expected_frequency", "10min")), source: { adapter: pythonString(defaultIfAbsent(source, "adapter", "synthetic")), mode: choice(defaultIfAbsent(source, "mode", "direct_column"), ["direct_column", "sql", "derived"], "parameter source mode"), source_metric: optionalString(source.source_metric, "source_metric"), source_group: optionalString(source.source_group, "source_group"), measurement: optionalString(source.measurement, "measurement"), field: optionalString(source.field, "field"), tags: stringMap(defaultIfAbsent(source, "tags", {}), "source tags"), schema: optionalString(aliasIfAbsent(source, "schema", "schema_name"), "schema"), table: optionalString(source.table, "table"), time_column: optionalString(source.time_column, "time_column"), value_column: optionalString(source.value_column, "value_column"), metric_name_column: optionalString(source.metric_name_column, "metric_name_column"), metric_name: optionalString(source.metric_name, "metric_name"), unit_column: optionalString(source.unit_column, "unit_column"), unit: optionalString(aliasIfAbsent(source, "unit", "source_unit"), "source unit"), filters: stringMap(defaultIfAbsent(source, "filters", {}), "source filters"), sql: optionalString(source.sql, "source sql"), params: object(defaultIfAbsent(source, "params", {}), "source params") }, time_semantics: { event_time_field: pythonString(defaultIfAbsent(time, "event_time_field", "event_time")), ingested_at_field: pythonString(defaultIfAbsent(time, "ingested_at_field", "ingested_at")), timezone: pythonString(defaultIfAbsent(time, "timezone", "Asia/Shanghai")), availability: normalizeAvailability(time.availability) }, availability_sla: normalizeAvailabilitySla(payload.availability_sla), value_field: pythonString(defaultIfAbsent(payload, "value_field", "value")), quality_rules: normalizeQualityRules(object(defaultIfAbsent(payload, "quality_rules", {}), "parameter quality_rules")), rounding: normalizeRounding(payload.rounding), owner: pythonString(defaultIfAbsent(payload, "owner", "demo")) };
 }
 
-function normalizeOperator(payload) { exactKeys(payload, ["schema_version", "project", "name", "version", "type", "function_hash", "entrypoint", "code_hash", "package_uri", "code_artifact", "input_schema", "output_schema", "config_schema", "runtime", "resources", "deterministic", "supports_batch", "supports_online", "owner"], ["name", "version"], "operator"); const artifact = payload.code_artifact == null ? null : normalizeCodeArtifact(payload.code_artifact); const normalized = { schema_version: pythonString(defaultIfAbsent(payload, "schema_version", "ml_data_platform.operator/v1")), project: identifier(defaultIfAbsent(payload, "project", "default"), "operator project"), name: identifier(payload.name, "operator name"), version: identifier(payload.version, "operator version"), type: choice(defaultIfAbsent(payload, "type", "standard"), ["standard", "custom_python", "spark_udf", "feature"], "operator type"), function_hash: pythonString(defaultIfAbsent(payload, "function_hash", "demo")), entrypoint: optionalString(payload.entrypoint, "operator entrypoint"), code_hash: optionalString(payload.code_hash, "operator code_hash"), package_uri: optionalString(payload.package_uri, "operator package_uri"), code_artifact: artifact, input_schema: object(defaultIfAbsent(payload, "input_schema", {}), "operator input_schema"), output_schema: object(defaultIfAbsent(payload, "output_schema", {}), "operator output_schema"), config_schema: object(defaultIfAbsent(payload, "config_schema", {}), "operator config_schema"), runtime: object(defaultIfAbsent(payload, "runtime", {}), "operator runtime"), resources: object(defaultIfAbsent(payload, "resources", {}), "operator resources"), deterministic: bool(defaultIfAbsent(payload, "deterministic", true), "operator deterministic"), supports_batch: bool(defaultIfAbsent(payload, "supports_batch", true), "operator supports_batch"), supports_online: bool(defaultIfAbsent(payload, "supports_online", false), "operator supports_online"), owner: pythonString(defaultIfAbsent(payload, "owner", "demo")) }; if (artifact && normalized.code_hash !== artifact.sha256) throw new Error("operator code_hash must equal code_artifact.sha256"); if (artifact && normalized.package_uri !== artifact.uri) throw new Error("operator package_uri must equal code_artifact.uri"); return normalized; }
+function normalizeOperator(payload) { object(payload, "operator"); const artifact = payload.code_artifact == null ? null : normalizeCodeArtifact(payload.code_artifact); const normalized = { schema_version: pythonString(defaultIfAbsent(payload, "schema_version", "ml_data_platform.operator/v1")), project: identifier(defaultIfAbsent(payload, "project", "default"), "operator project"), name: identifier(payload.name, "operator name"), version: identifier(payload.version, "operator version"), type: choice(defaultIfAbsent(payload, "type", "standard"), ["standard", "custom_python", "spark_udf", "feature"], "operator type"), function_hash: pythonString(defaultIfAbsent(payload, "function_hash", "demo")), entrypoint: optionalString(payload.entrypoint, "operator entrypoint"), code_hash: optionalString(payload.code_hash, "operator code_hash"), package_uri: optionalString(payload.package_uri, "operator package_uri"), code_artifact: artifact, input_schema: object(defaultIfAbsent(payload, "input_schema", {}), "operator input_schema"), output_schema: object(defaultIfAbsent(payload, "output_schema", {}), "operator output_schema"), config_schema: object(defaultIfAbsent(payload, "config_schema", {}), "operator config_schema"), runtime: object(defaultIfAbsent(payload, "runtime", {}), "operator runtime"), resources: object(defaultIfAbsent(payload, "resources", {}), "operator resources"), deterministic: bool(defaultIfAbsent(payload, "deterministic", true), "operator deterministic"), supports_batch: bool(defaultIfAbsent(payload, "supports_batch", true), "operator supports_batch"), supports_online: bool(defaultIfAbsent(payload, "supports_online", false), "operator supports_online"), owner: pythonString(defaultIfAbsent(payload, "owner", "demo")) }; if (artifact && normalized.code_hash !== artifact.sha256) throw new Error("operator code_hash must equal code_artifact.sha256"); if (artifact && normalized.package_uri !== artifact.uri) throw new Error("operator package_uri must equal code_artifact.uri"); return normalized; }
 function normalizeRegistryOperator(payload) { const value = object(payload, "operator"); if (!hasOwn(value, "deleted_at")) return normalizeOperator(value); const { deleted_at, ...contract } = value; void deleted_at; return normalizeOperator(contract); }
 function normalizeFeature(payload) { exactKeys(payload, ["schema_version", "project", "name", "version", "inputs", "operator", "operator_version", "config", "output_column", "output_dtype", "offline_online_supported", "owner", "description"], ["name", "version", "inputs", "operator", "operator_version", "output_column"], "feature"); const project = identifier(defaultIfAbsent(payload, "project", "default"), "feature project"); const inputs = objectList(payload.inputs, "feature inputs").map((item) => parameterRef(item, project)); if (!inputs.length) throw new Error("feature inputs must not be empty"); unique(inputs.map(parameterReferenceKey), "feature inputs"); if (typeof payload.output_column !== "string" || !outputColumnPattern.test(payload.output_column)) throw new Error("feature output_column is invalid"); return { schema_version: pythonString(defaultIfAbsent(payload, "schema_version", "ml_data_platform.feature/v1")), project, name: identifier(payload.name, "feature name"), version: identifier(payload.version, "feature version"), inputs, operator: identifier(payload.operator, "feature operator"), operator_version: identifier(payload.operator_version, "feature operator_version"), config: object(defaultIfAbsent(payload, "config", {}), "feature config"), output_column: payload.output_column, output_dtype: pythonString(defaultIfAbsent(payload, "output_dtype", "float64")), offline_online_supported: bool(defaultIfAbsent(payload, "offline_online_supported", true), "offline_online_supported"), owner: pythonString(defaultIfAbsent(payload, "owner", "demo")), description: pythonString(defaultIfAbsent(payload, "description", "")) }; }
 function normalizeFeatureSet(payload) { exactKeys(payload, ["schema_version", "project", "name", "version", "features", "owner", "description"], ["name", "version", "features"], "feature set"); const features = objectList(payload.features, "feature set features").map((item) => ref(item, "feature reference")); if (!features.length) throw new Error("feature_set features must not be empty"); unique(features.map(referenceKey), "feature_set features"); return { schema_version: pythonString(defaultIfAbsent(payload, "schema_version", "ml_data_platform.feature_set/v1")), project: identifier(defaultIfAbsent(payload, "project", "default"), "feature set project"), name: identifier(payload.name, "feature set name"), version: identifier(payload.version, "feature set version"), features, owner: pythonString(defaultIfAbsent(payload, "owner", "demo")), description: pythonString(defaultIfAbsent(payload, "description", "")) }; }

package/runtime/business-client/src/cli.js

@@ -15,11 +15,21 @@ const BUSINESS_COMMANDS = new Set([
   "list-feature-sets", "list-datasets", "get-dataset", "list-jobs", "resolve-manifest",
   "resolve-dataset", "build-dataset", "build-registered-dataset",
   "get-parameter", "get-feature", "get-feature-set", "get-operator",
+  "delete-parameter", "delete-feature", "delete-feature-set",
+  "restore-parameter", "restore-feature", "restore-feature-set",
+  "get-parameter-lifecycle", "get-feature-lifecycle", "get-feature-set-lifecycle",
+  "list-parameter-references", "list-feature-references", "list-feature-set-references",
   "list-parameter-versions", "list-feature-versions", "list-feature-set-versions", "list-operator-versions",
   "add-parameter", "add-feature", "add-feature-set", "add-operator", "publish-operator",
   "add-dataset", "update-dataset", "delete-dataset", "list-dataset-artifacts",
   "register-operator", "show-operator-specs", "delete-operator", "run-operator",
-  "get-job", "wait-job", "retry-job", "cancel-job", "get-dataset-artifact", "download-dataset-artifact", "fetch-inference-data", "apply",
+  "get-job", "wait-job", "retry-job", "cancel-job", "get-dataset-artifact", "download-dataset-artifact", "fetch-inference-data", "fetch-inference-context", "apply",
+  "register-trainer", "list-trainer-definitions", "get-trainer-definition", "register-training-run", "validate-training-run", "submit-training-job", "get-training-job", "retry-training-job", "cancel-training-job",
+  "register-evaluation-config", "get-evaluation-config", "register-metric-definition", "get-metric-definition",
+  "register-executable-package", "verify-executable-package", "publish-executable-package",
+  "validate-evaluation-run", "submit-evaluation", "get-evaluation-run", "get-evaluation-job", "retry-evaluation-job", "cancel-evaluation-job", "list-evaluation-attempts",
+  "get-evaluation-result", "get-evaluation-result-metrics", "get-evaluation-result-artifacts", "summarize-evaluation-run", "get-evaluation-summary",
+  "create-model-package", "get-model-package", "cancel-model-package", "validate-model-artifact",
 ]);
 
 const COMMAND_USAGE = {
@@ -40,6 +50,18 @@ const COMMAND_USAGE = {
   "get-feature": "get-feature NAME VERSION [--project PROJECT]",
   "get-feature-set": "get-feature-set NAME VERSION [--project PROJECT]",
   "get-operator": "get-operator NAME VERSION [--project PROJECT]",
+  "delete-parameter": "delete-parameter NAME VERSION [--project PROJECT] [--actor ACTOR] [--reason TEXT] [--idempotency-key KEY]",
+  "delete-feature": "delete-feature NAME VERSION [--project PROJECT] [--actor ACTOR] [--reason TEXT] [--idempotency-key KEY]",
+  "delete-feature-set": "delete-feature-set NAME VERSION [--project PROJECT] [--actor ACTOR] [--reason TEXT] [--idempotency-key KEY]",
+  "restore-parameter": "restore-parameter NAME VERSION [--project PROJECT] [--actor ACTOR] [--reason TEXT] [--idempotency-key KEY]",
+  "restore-feature": "restore-feature NAME VERSION [--project PROJECT] [--actor ACTOR] [--reason TEXT] [--idempotency-key KEY]",
+  "restore-feature-set": "restore-feature-set NAME VERSION [--project PROJECT] [--actor ACTOR] [--reason TEXT] [--idempotency-key KEY]",
+  "get-parameter-lifecycle": "get-parameter-lifecycle NAME VERSION [--project PROJECT]",
+  "get-feature-lifecycle": "get-feature-lifecycle NAME VERSION [--project PROJECT]",
+  "get-feature-set-lifecycle": "get-feature-set-lifecycle NAME VERSION [--project PROJECT]",
+  "list-parameter-references": "list-parameter-references NAME VERSION [--project PROJECT] [--direct]",
+  "list-feature-references": "list-feature-references NAME VERSION [--project PROJECT] [--direct]",
+  "list-feature-set-references": "list-feature-set-references NAME VERSION [--project PROJECT] [--direct]",
   "list-parameter-versions": "list-parameter-versions NAME [--project PROJECT] [--limit N] [--offset N] [--all]",
   "list-feature-versions": "list-feature-versions NAME [--project PROJECT] [--limit N] [--offset N] [--all]",
   "list-feature-set-versions": "list-feature-set-versions NAME [--project PROJECT] [--limit N] [--offset N] [--all]",
@@ -70,6 +92,39 @@ const COMMAND_USAGE = {
   "get-dataset-artifact": "get-dataset-artifact DATASET_ID MANIFEST_HASH [--project PROJECT]",
   "download-dataset-artifact": "download-dataset-artifact DATASET_ID MANIFEST_HASH [--project PROJECT] --out-dir PATH [--file PATH] [--force]",
   "fetch-inference-data": "fetch-inference-data MANIFEST --cutoff-time TIMESTAMP [--max-workers N] [--max-source-lag-hours HOURS] [--no-validate-freshness] [--allow-missing]",
+  "fetch-inference-context": "fetch-inference-context REQUEST_JSON",
+  "register-trainer": "register-trainer DEFINITION_JSON",
+  "list-trainer-definitions": "list-trainer-definitions",
+  "get-trainer-definition": "get-trainer-definition NAME VERSION [--project PROJECT]",
+  "register-training-run": "register-training-run RUN_JSON",
+  "validate-training-run": "validate-training-run RUN_JSON",
+  "submit-training-job": "submit-training-job REQUEST_JSON",
+  "get-training-job": "get-training-job JOB_ID",
+  "retry-training-job": "retry-training-job JOB_ID",
+  "cancel-training-job": "cancel-training-job JOB_ID",
+  "register-evaluation-config": "register-evaluation-config CONFIG_JSON",
+  "get-evaluation-config": "get-evaluation-config NAME VERSION",
+  "register-metric-definition": "register-metric-definition DEFINITION_JSON",
+  "get-metric-definition": "get-metric-definition NAME VERSION",
+  "register-executable-package": "register-executable-package SPEC_JSON",
+  "verify-executable-package": "verify-executable-package NAME VERSION",
+  "publish-executable-package": "publish-executable-package NAME VERSION",
+  "validate-evaluation-run": "validate-evaluation-run REQUEST_JSON",
+  "submit-evaluation": "submit-evaluation REQUEST_JSON",
+  "get-evaluation-run": "get-evaluation-run RUN_ID",
+  "get-evaluation-job": "get-evaluation-job JOB_ID",
+  "retry-evaluation-job": "retry-evaluation-job JOB_ID",
+  "cancel-evaluation-job": "cancel-evaluation-job JOB_ID",
+  "list-evaluation-attempts": "list-evaluation-attempts JOB_ID",
+  "get-evaluation-result": "get-evaluation-result RESULT_ID",
+  "get-evaluation-result-metrics": "get-evaluation-result-metrics RESULT_ID",
+  "get-evaluation-result-artifacts": "get-evaluation-result-artifacts RESULT_ID",
+  "summarize-evaluation-run": "summarize-evaluation-run RUN_ID",
+  "get-evaluation-summary": "get-evaluation-summary SUMMARY_ID",
+  "create-model-package": "create-model-package REQUEST_JSON",
+  "get-model-package": "get-model-package PACKAGE_ID [--package-version VERSION]",
+  "cancel-model-package": "cancel-model-package PACKAGE_ID [--package-version VERSION]",
+  "validate-model-artifact": "validate-model-artifact ARTIFACT_ID",
   apply: "apply CATALOG_DIRECTORY [--dry-run] [--project PROJECT]",
 };
 
@@ -153,6 +208,17 @@ function datasetEndpoint(project, datasetId, datasetVersion) {
   return `/datasets/${encodeURIComponent(project)}/${encodeURIComponent(datasetId)}/${encodeURIComponent(datasetVersion)}`;
 }
 
+function lifecycleResource(command) {
+  if (command.includes("feature-set")) return "feature-sets";
+  if (command.includes("parameter")) return "parameters";
+  if (command.includes("feature")) return "features";
+  throw new Error(`unsupported lifecycle command: ${command}`);
+}
+
+function lifecycleEndpoint(resource, project, name, version, suffix = "") {
+  return `/registry/${resource}/${encodeURIComponent(project)}/${encodeURIComponent(name)}/${encodeURIComponent(version)}${suffix}`;
+}
+
 async function registeredDataset(api, project, datasetId, datasetVersion, includeDeleted = false) {
   const payload = await api.get(datasetEndpoint(project, datasetId, datasetVersion), { include_deleted: includeDeleted });
   if (!payload || typeof payload !== "object" || Array.isArray(payload)) throw new Error("registered dataset response must be a JSON object");
@@ -262,6 +328,22 @@ async function runBusinessCli(argv) {
     const version = positional(rest, `${resource} version`);
     const project = take(rest, "--project", "default");
     result = await client(options).get(`/${endpoint}/${encodeURIComponent(project)}/${encodeURIComponent(name)}/${encodeURIComponent(version)}`);
+  } else if (/^delete-(parameter|feature|feature-set)$/.test(options.command)) {
+    const name = positional(rest, "resource name"); const version = positional(rest, "resource version"); const project = take(rest, "--project", "default");
+    result = await client(options).delete(lifecycleEndpoint(lifecycleResource(options.command), project, name, version), {
+      actor: take(rest, "--actor"), reason: take(rest, "--reason"), idempotency_key: take(rest, "--idempotency-key"),
+    });
+  } else if (/^restore-(parameter|feature|feature-set)$/.test(options.command)) {
+    const name = positional(rest, "resource name"); const version = positional(rest, "resource version"); const project = take(rest, "--project", "default");
+    result = await client(options).post(lifecycleEndpoint(lifecycleResource(options.command), project, name, version, "/restore"), {
+      actor: take(rest, "--actor") || "anonymous", reason: take(rest, "--reason", ""), idempotency_key: take(rest, "--idempotency-key"),
+    });
+  } else if (/^get-(parameter|feature|feature-set)-lifecycle$/.test(options.command)) {
+    const name = positional(rest, "resource name"); const version = positional(rest, "resource version"); const project = take(rest, "--project", "default");
+    result = await client(options).get(lifecycleEndpoint(lifecycleResource(options.command), project, name, version, "/lifecycle"));
+  } else if (/^list-(parameter|feature|feature-set)-references$/.test(options.command)) {
+    const name = positional(rest, "resource name"); const version = positional(rest, "resource version"); const project = take(rest, "--project", "default");
+    result = await client(options).get(lifecycleEndpoint(lifecycleResource(options.command), project, name, version, "/references"), { transitive: !boolean(rest, "--direct") });
   } else if (["list-parameter-versions", "list-feature-versions", "list-feature-set-versions", "list-operator-versions"].includes(options.command)) {
     const resource = options.command.replace(/^list-/, "").replace(/-versions$/, "");
     const endpoint = resource === "feature-set" ? "feature-sets" : `${resource}s`;
@@ -272,7 +354,7 @@ async function runBusinessCli(argv) {
     if (allPages && (rawLimit !== undefined || rawOffset !== undefined)) throw new Error("--all cannot be combined with --limit or --offset");
     if (allPages) result = await fetchAll(client(options), `/${endpoint}/${encodeURIComponent(project)}/${encodeURIComponent(name)}`);
     else { const limit = Number(rawLimit ?? 50); const offset = Number(rawOffset ?? 0); if (!Number.isInteger(limit) || limit < 1 || limit > 500) throw new Error("limit must be between 1 and 500"); if (!Number.isInteger(offset) || offset < 0) throw new Error("offset must be >= 0"); result = await client(options).get(`/${endpoint}/${encodeURIComponent(project)}/${encodeURIComponent(name)}`, { limit, offset }); }
-  } else if (options.command.startsWith("list-") && !["list-jobs", "list-dataset-artifacts"].includes(options.command)) {
+  } else if (options.command.startsWith("list-") && !["list-jobs", "list-dataset-artifacts", "list-trainer-definitions"].includes(options.command)) {
     const endpoints = { "list-parameters": "/parameters", "list-operators": "/operators", "list-features": "/features", "list-feature-sets": "/feature-sets", "list-datasets": "/datasets" };
     const project = take(rest, "--project");
     const name = take(rest, "--name"); const version = take(rest, "--version");
@@ -355,6 +437,45 @@ async function runBusinessCli(argv) {
     const manifest = positional(rest, "manifest"); const cutoffTime = take(rest, "--cutoff-time"); if (!cutoffTime) throw new Error("--cutoff-time is required");
     const realtime = {}; const maxWorkers = take(rest, "--max-workers"); const maxLag = take(rest, "--max-source-lag-hours"); if (maxWorkers !== undefined) { realtime.max_workers = integer(maxWorkers, "--max-workers"); if (realtime.max_workers < 1) throw new Error("--max-workers must be at least 1"); } if (maxLag !== undefined) { realtime.max_source_lag_hours = finiteNumber(maxLag, "--max-source-lag-hours"); if (realtime.max_source_lag_hours < 0) throw new Error("--max-source-lag-hours must be non-negative"); } if (boolean(rest, "--no-validate-freshness")) realtime.validate_freshness = false; if (boolean(rest, "--allow-missing")) realtime.allow_missing = true;
     const request = { manifest: readDataset(manifest), cutoff_time: cutoffTime }; if (Object.keys(realtime).length) request.realtime_fetch = realtime; result = await client(options).post("/inference-data/fetch", request);
+  } else if (options.command === "fetch-inference-context") {
+    result = await client(options).post("/inference-data/context", readSpec(positional(rest, "request JSON")));
+  } else if (options.command === "list-trainer-definitions") {
+    result = await client(options).get("/trainer-definitions");
+  } else if (options.command === "get-trainer-definition") {
+    const name = positional(rest, "trainer name"); const version = positional(rest, "trainer version"); const project = take(rest, "--project", "default");
+    result = await client(options).get(`/trainer-definitions/${encodeURIComponent(project)}/${encodeURIComponent(name)}/${encodeURIComponent(version)}`);
+  } else if ({
+    "register-trainer": "/trainer-definitions", "register-training-run": "/training-runs", "validate-training-run": "/training-runs/validate",
+    "submit-training-job": "/training-jobs", "register-evaluation-config": "/evaluation-configs", "register-metric-definition": "/metric-definitions",
+    "register-executable-package": "/executable-packages", "validate-evaluation-run": "/evaluation-runs/validate", "submit-evaluation": "/evaluation-runs",
+    "create-model-package": "/model-packages",
+  }[options.command]) {
+    const endpoint = { "register-trainer": "/trainer-definitions", "register-training-run": "/training-runs", "validate-training-run": "/training-runs/validate", "submit-training-job": "/training-jobs", "register-evaluation-config": "/evaluation-configs", "register-metric-definition": "/metric-definitions", "register-executable-package": "/executable-packages", "validate-evaluation-run": "/evaluation-runs/validate", "submit-evaluation": "/evaluation-runs", "create-model-package": "/model-packages" }[options.command];
+    result = await client(options).post(endpoint, readSpec(positional(rest, "request JSON")));
+  } else if (/^(get|retry|cancel)-training-job$/.test(options.command)) {
+    const jobId = positional(rest, "training job id"); const endpoint = `/training-jobs/${encodeURIComponent(jobId)}`;
+    result = options.command === "get-training-job" ? await client(options).get(endpoint) : await client(options).post(`${endpoint}/${options.command.split("-")[0]}`);
+  } else if (/^(get|retry|cancel)-evaluation-job$/.test(options.command) || options.command === "list-evaluation-attempts") {
+    const jobId = positional(rest, "evaluation job id"); const endpoint = `/evaluation-jobs/${encodeURIComponent(jobId)}`;
+    if (options.command === "get-evaluation-job") result = await client(options).get(endpoint);
+    else if (options.command === "list-evaluation-attempts") result = await client(options).get(`${endpoint}/attempts`);
+    else result = await client(options).post(`${endpoint}/${options.command.split("-")[0]}`);
+  } else if (options.command === "get-evaluation-run") result = await client(options).get(`/evaluation-runs/${encodeURIComponent(positional(rest, "evaluation run id"))}`);
+  else if (options.command === "summarize-evaluation-run") result = await client(options).post(`/evaluation-runs/${encodeURIComponent(positional(rest, "evaluation run id"))}/summarize`);
+  else if (options.command === "get-evaluation-summary") result = await client(options).get(`/evaluation-summaries/${encodeURIComponent(positional(rest, "evaluation summary id"))}`);
+  else if (/^get-evaluation-result(?:-metrics|-artifacts)?$/.test(options.command)) {
+    const resultId = positional(rest, "evaluation result id"); const suffix = options.command.endsWith("-metrics") ? "/metrics" : options.command.endsWith("-artifacts") ? "/artifacts" : "";
+    result = await client(options).get(`/evaluation-results/${encodeURIComponent(resultId)}${suffix}`);
+  } else if (/^(get-evaluation-config|get-metric-definition)$/.test(options.command)) {
+    const name = positional(rest, "resource name"); const version = positional(rest, "resource version"); const base = options.command === "get-evaluation-config" ? "evaluation-configs" : "metric-definitions";
+    result = await client(options).get(`/${base}/${encodeURIComponent(name)}/${encodeURIComponent(version)}`);
+  } else if (/^(verify|publish)-executable-package$/.test(options.command)) {
+    const name = positional(rest, "package name"); const version = positional(rest, "package version"); const action = options.command.split("-")[0];
+    result = await client(options).post(`/executable-packages/${encodeURIComponent(name)}/${encodeURIComponent(version)}/${action}`);
+  } else if (options.command === "validate-model-artifact") result = await client(options).get(`/model-artifacts/${encodeURIComponent(positional(rest, "model artifact id"))}/validate`);
+  else if (/^(get|cancel)-model-package$/.test(options.command)) {
+    const packageId = positional(rest, "model package id"); const params = { package_version: take(rest, "--package-version") }; const endpoint = `/model-packages/${encodeURIComponent(packageId)}`;
+    result = options.command === "get-model-package" ? await client(options).get(endpoint, params) : await client(options).post(`${endpoint}/cancel`, undefined, params);
   } else if (options.command === "apply") {
     const directory = positional(rest, "catalog directory"); const dryRun = boolean(rest, "--dry-run"); const targetProject = take(rest, "--project");
     result = await applyCatalog(loadCatalog(resolvedPath(directory)), client(options), dryRun, targetProject);

package/scripts/lib.js

@@ -152,19 +152,21 @@ function loadRelease(packageRoot = PACKAGE_ROOT) {
       || requirements.os.join(",") !== "darwin,linux") {
     throw new Error("release manifest 缺少有效 runtime requirements");
   }
-  const skill = release.skills && release.skills[SKILL_NAME];
-  if (!skill || skill.path !== `skills/${SKILL_NAME}` || !isSha256(skill.sha256)
-      || !isSemver(skill.revision) || typeof skill.requires_cli !== "string") {
+  const skills = release.skills;
+  const skill = skills && skills[SKILL_NAME];
+  if (!skill || Object.keys(skills).length < 1) {
     throw new Error("release manifest 缺少有效 Skill 信息");
   }
+  for (const [name, item] of Object.entries(skills)) {
+    if (item.path !== `skills/${name}` || !isSha256(item.sha256)
+        || !isSemver(item.revision) || typeof item.requires_cli !== "string") throw new Error("release manifest 缺少有效 Skill 信息");
+  }
   const cli = release.cli;
   if (!cli || cli.name !== CLI_NAME || cli.path !== "runtime/business-client"
       || !isSemver(cli.version) || !isSha256(cli.sha256) || cli.entrypoint !== "src/cli.js") {
     throw new Error("release manifest 缺少有效 JavaScript CLI 信息");
   }
-  if (!versionSatisfies(cli.version, skill.requires_cli)) {
-    throw new Error("Skill requires_cli 与 release CLI version 不兼容");
-  }
+  for (const item of Object.values(skills)) if (!versionSatisfies(cli.version, item.requires_cli)) throw new Error("Skill requires_cli 与 release CLI version 不兼容");
   return release;
 }
 
@@ -201,7 +203,7 @@ function verifyPackage(packageRoot = PACKAGE_ROOT) {
   }
 
   assertTopLevel(path.join(packageRoot, "runtime"), new Set(["business-client"]), "runtime");
-  assertTopLevel(path.join(packageRoot, "skills"), new Set([SKILL_NAME]), "skills");
+  assertTopLevel(path.join(packageRoot, "skills"), new Set(Object.keys(release.skills)), "skills");
   const skillRoot = resolveInside(packageRoot, release.skills[SKILL_NAME].path);
   const clientRoot = resolveInside(packageRoot, release.cli.path);
   assertTopLevel(skillRoot, SKILL_TOP_LEVEL, "Skill");
@@ -232,11 +234,18 @@ function verifyPackage(packageRoot = PACKAGE_ROOT) {
   if (treeDigest(skillRoot) !== release.skills[SKILL_NAME].sha256) {
     throw new Error("Skill 目录摘要与 release manifest 不一致");
   }
+  for (const [name, item] of Object.entries(release.skills)) {
+    const root = resolveInside(packageRoot, item.path);
+    const actual = new Set(fs.readdirSync(root));
+    if (!actual.has("SKILL.md") || !actual.has("agents") || [...actual].some((entry) => !SKILL_TOP_LEVEL.has(entry))) throw new Error(`Skill ${name} 目录结构无效`);
+    rejectForbiddenTree(root, `Skill ${name}`);
+    if (treeDigest(root) !== item.sha256) throw new Error(`Skill ${name} 目录摘要与 release manifest 不一致`);
+  }
   if (treeDigest(clientRoot) !== release.cli.sha256) {
     throw new Error("JavaScript CLI 摘要与 release manifest 不一致");
   }
   if (!fs.existsSync(expectedEntry)) throw new Error("JavaScript CLI 入口不存在");
-  return { release, skillRoot, clientRoot };
+  return { release, skillRoot, skillRoots: Object.fromEntries(Object.entries(release.skills).map(([name, item]) => [name, resolveInside(packageRoot, item.path)])), clientRoot };
 }
 
 function status(options = {}, packageRoot = PACKAGE_ROOT) {
@@ -516,6 +525,9 @@ function installLocked(options, packageRoot, verified, paths) {
   let hadShim = false;
   let legacyBackupPath = null;
   let legacyMoved = false;
+  const additionalSkills = Object.keys(verified.release.skills).filter((name) => name !== SKILL_NAME);
+  const additionalBackups = [];
+  const installedAdditional = [];
   try {
     stagePersistentInstall(stagedInstall, packageRoot, verified);
     fs.cpSync(verified.skillRoot, stagedSkill, { recursive: true, errorOnExist: true });
@@ -540,6 +552,17 @@ function installLocked(options, packageRoot, verified, paths) {
     }
     fs.renameSync(stagedSkill, paths.target);
     skillMoved = true;
+    for (const name of additionalSkills) {
+      const target = path.join(paths.skillsDir, name);
+      const backup = path.join(paths.skillsDir, `.${name}.backup.${process.pid}.${Date.now()}`);
+      if (pathEntryExists(target)) {
+        if (!current?.skills?.[name] || current.skills[name].path !== target) throw new Error(`拒绝覆盖未托管 Skill: ${name}`);
+        fs.renameSync(target, backup);
+        additionalBackups.push({ target, backup });
+      }
+      fs.cpSync(verified.skillRoots[name], target, { recursive: true, errorOnExist: true });
+      installedAdditional.push(target);
+    }
     if (pathEntryExists(paths.shim)) {
       fs.renameSync(paths.shim, shimBackup);
       hadShim = true;
@@ -555,6 +578,7 @@ function installLocked(options, packageRoot, verified, paths) {
     removeOwnedDirectory(skillStaging, paths.skillsDir, `.${SKILL_NAME}.install.`);
     if (hadState) removeOwnedDirectory(stateBackup, paths.stateDir, ".backup.");
     if (hadSkill) removeOwnedDirectory(skillBackup, paths.skillsDir, `.${SKILL_NAME}.backup.`);
+    for (const { backup } of additionalBackups) fs.rmSync(backup, { recursive: true, force: true });
     if (hadShim) fs.rmSync(shimBackup, { force: true });
     const result = installationResult(
       options.migrateLegacy ? "migrated" : (current || (targetExists && !backingUpUnmanaged) ? "upgraded" : "installed"),
@@ -572,6 +596,8 @@ function installLocked(options, packageRoot, verified, paths) {
     if (hadShim && pathEntryExists(shimBackup)) fs.renameSync(shimBackup, paths.shim);
     if (skillMoved && pathEntryExists(paths.target)) removeOwnedDirectory(paths.target, paths.skillsDir, SKILL_NAME);
     if (hadSkill && pathEntryExists(skillBackup)) fs.renameSync(skillBackup, paths.target);
+    for (const target of installedAdditional.reverse()) fs.rmSync(target, { recursive: true, force: true });
+    for (const { target, backup } of additionalBackups.reverse()) if (pathEntryExists(backup)) fs.renameSync(backup, target);
     if (legacyMoved && legacyBackupPath && pathEntryExists(legacyBackupPath) && !pathEntryExists(paths.target)) {
       fs.renameSync(legacyBackupPath, paths.target);
     }
@@ -613,7 +639,7 @@ function stagePersistentInstall(target, packageRoot, verified) {
 }
 
 function buildInstallationRecord(paths, release) {
-  return {
+  const record = {
     schema_version: RECORD_SCHEMA,
     owner: { package: PACKAGE_NAME, installation_id: installationId(paths) },
     release_version: release.release_version,
@@ -637,11 +663,17 @@ function buildInstallationRecord(paths, release) {
     release_sha256: sha256File(path.join(paths.installRoot, "release.json")),
     policy_sha256: sha256File(path.join(paths.installRoot, "release-policy.json")),
   };
+  record.skills = Object.fromEntries(Object.entries(release.skills).map(([name, item]) => [name, {
+    name, revision: item.revision, requires_cli: item.requires_cli,
+    path: path.join(paths.skillsDir, name), sha256: item.sha256,
+  }]));
+  return record;
 }
 
 function assertInstalledState(paths, record, prewarm) {
   if (!digestEquals(paths.clientRoot, record.cli.sha256)) throw new Error("安装后的 CLI 摘要不匹配");
   if (!digestEquals(paths.target, record.skill.sha256)) throw new Error("安装后的 Skill 摘要不匹配");
+  for (const skill of Object.values(record.skills || {})) if (!digestEquals(skill.path, skill.sha256)) throw new Error(`安装后的 Skill 摘要不匹配: ${skill.name}`);
   if (sha256File(paths.shim) !== record.shim.sha256) throw new Error("安装后的 CLI shim 摘要不匹配");
   if (prewarm) assertRuntime(paths.dispatcher);
 }

package/skills/feature-management/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: feature-management
-description: Manage projects, versioned business feature assets, dataset rowsets and endpoint eligibility, and causal realtime feature reads through the deployed ML Platform API using the ml-platform CLI installed with the same release. Use when a business user asks an agent to manage an empty Project, define or change Parameter source contracts, author/package/register an Operator wheel, define single-column Features, order a FeatureSet, create a DatasetManifest or catalog, configure training/validation/test rowsets or abnormal-data endpoint exclusion, configure realtime freshness policy, run catalog dry-run/apply, build a dataset, fetch one inference row, or inspect an artifact without downloading the platform source repository. Do not use for model training, versioned Registry asset deletion, image builds, or Kubernetes deployment.
+description: Manage projects, versioned business feature assets and their lifecycle, dataset rowsets and endpoint eligibility, and causal realtime row or sequence-context reads through the deployed ML Platform API using the ml-platform CLI installed with the same release. Use when a business user asks an agent to manage an empty Project, define or change Parameter source contracts, author/package/register an Operator wheel, define single-column Features, order a FeatureSet, create a DatasetManifest or catalog, inspect references, soft-delete or restore a Parameter/Feature/FeatureSet version, configure training/validation/test rowsets or abnormal-data endpoint exclusion, configure realtime freshness policy, run catalog dry-run/apply, build a dataset, fetch inference data, or inspect an artifact without downloading the platform source repository. Do not use for model training, Operator/Dataset deletion, image builds, or Kubernetes deployment.
 ---
 
 # Feature Management
@@ -117,6 +117,16 @@ Confirm:
 
 Separate preparation from mutation. Creating files and running local tests or `apply --dry-run` does not authorize publishing a wheel, changing the Registry, or submitting a build Job.
 
+Soft-delete and restore are separate Registry mutations. Before soft-deleting a
+Parameter, Feature, or FeatureSet version, inspect its lifecycle state and both
+direct and transitive references, report the affected immutable consumers, and
+obtain explicit authorization for the exact `project/name:version`. Supply an
+actor, reason, and stable idempotency key. Soft deletion preserves historical
+references and hashes but hides the version from normal discovery and prevents
+new consumers from binding it. Restore only after checking that all of the
+resource's own dependencies are active. Never describe these operations as
+physical deletion.
+
 A label such as "10-minute mean" confirms neither its input Parameter nor a
 complete formula. Before drafting an Operator or Feature, ask for the exact
 versioned input Parameter(s), or for an approved source contract from which

package/skills/feature-management/references/commands.md

@@ -10,8 +10,9 @@ Use the `ml-platform` executable installed on `PATH` by the same ML Platform rel
 4. Validate and publish a catalog
 5. Resolve and build a dataset
 6. Fetch realtime inference data
-7. Inspect and download an artifact
-8. Prohibited shortcuts
+7. Inspect and change Registry lifecycle state
+8. Inspect and download an artifact
+9. Prohibited shortcuts
 
 ## Select A Target
 
@@ -257,6 +258,57 @@ write parquet, or publish a DatasetArtifact. Treat HTTP `503` as retryable sourc
 freshness/missing-data failure and HTTP `400` as a contract error that requires a
 manifest or request change.
 
+For a deployment-bound row or full sequence context, prepare one JSON object
+containing `feature_retrieval`, `cutoff_time`, and `entity_keys`, then run:
+
+```bash
+ml-platform --profile server fetch-inference-context /tmp/inference-context.json
+```
+
+The `FeatureRetrievalSpec` must bind an existing inference-mode DatasetManifest
+by project, dataset id, version, and manifest hash; its FeatureSet and contract
+signature must match the resolved online contract. Context mode also fixes
+`context_length`, `frequency`, `context_end`, `stride`, input-adapter hash, and
+temporal-binding hash. Do not replace these immutable deployment bindings with
+ad-hoc CLI flags. As with row fetch, this operation writes no artifact.
+
+## Inspect And Change Registry Lifecycle State
+
+Lifecycle commands apply only to Parameter, Feature, and FeatureSet versions.
+Before a mutation, inspect the exact resource and its references:
+
+```bash
+ml-platform --profile server get-parameter-lifecycle <name> <version> --project <project>
+ml-platform --profile server list-parameter-references <name> <version> --project <project> --direct
+ml-platform --profile server list-parameter-references <name> <version> --project <project>
+```
+
+Equivalent `get-*-lifecycle` and `list-*-references` commands exist for
+`feature` and `feature-set`. The default reference query returns the transitive
+closure; `--direct` returns only immediate dependents.
+
+After explicit authorization for the exact resource, soft-delete it with
+auditable metadata:
+
+```bash
+ml-platform --profile server delete-parameter <name> <version> \
+  --project <project> --actor <actor> --reason '<reason>' \
+  --idempotency-key <stable-key>
+```
+
+Restore uses the same audit fields and is rejected if the resource's own
+dependencies are not active:
+
+```bash
+ml-platform --profile server restore-parameter <name> <version> \
+  --project <project> --actor <actor> --reason '<reason>' \
+  --idempotency-key <stable-key>
+```
+
+These operations never erase immutable specs, historical references, or
+contract hashes. Do not use the Operator or Dataset deletion commands as a
+substitute; they have separate lifecycle rules.
+
 ## Inspect And Download An Artifact
 
 ```bash
@@ -299,7 +351,7 @@ Use a structured Parquet reader for schema, row count, column order, and missing
 
 - Do not use `seed-demo v95` for business catalogs.
 - Do not call individual `add-*` commands when catalog `apply` is available for an end-to-end publication.
-- Do not use `delete-dataset`, `cancel-job`, or force overwrite operations unless separately requested.
+- Do not use any `delete-*`, `restore-*`, `cancel-job`, or force overwrite operation unless separately requested.
 - Do not build images, deploy Kubernetes, or change service configuration in this workflow.
 - Do not claim a successful publication from `--dry-run` output.
 - Do not claim Chronon compile/backfill from a successful first-phase Operator build.

package/skills/model-lifecycle-management/SKILL.md

@@ -0,0 +1,27 @@
+---
+name: model-lifecycle-management
+description: Manage ITSMP model training, evaluation, and immutable model packaging through the deployed platform API. Use when a user asks to list or inspect registered TrainerDefinitions, register a trainer, validate or submit a training run, inspect/retry/cancel training jobs, configure or run governed evaluation, inspect evaluation evidence, validate a ModelArtifact, or create and inspect a ModelPackage. Do not use for feature catalog authoring, dataset builds, model deployment, Kubernetes administration, or image release engineering.
+---
+
+# Model Lifecycle Management
+
+Use the `ml-platform` executable installed with the same npm release as this Skill. Verify `ml-platform version`, `ml-platform show-config`, and server health before API operations. Never use a platform source checkout or a second client runtime.
+
+```text
+DatasetArtifact -> TrainingRun -> TrainingJob -> ModelArtifact
+  -> EvaluationRun/Job -> EvaluationResult/Summary -> ModelPackage
+```
+
+Read [references/training.md](references/training.md) for training work, [references/evaluation.md](references/evaluation.md) for governed evaluation, and [references/packaging.md](references/packaging.md) for ModelArtifact validation and packaging. Read only the references required by the request.
+
+## Shared Gates
+
+- Establish exact immutable inputs, TrainerDefinition `v1`, runtime image digest, model inputs/targets, resources, evaluation policy, and package destination from authoritative contracts. Do not infer them from model names.
+- Validate, register, submit, retry, cancel, and package creation are separate actions. Read-only discovery and validation do not authorize mutation or workload submission.
+- Before submission, show the exact JSON path, immutable identities, expected workload, and target API. Obtain explicit authorization. Retry and cancel require separate authorization for the exact Job.
+- Never use `latest` image tags. During development TrainerDefinition stays at `v1`; rebinding it follows the reviewed release procedure and requires confirming no active Job references the old definition.
+- A succeeded Job is insufficient evidence. Training requires a valid ModelArtifact and signature; evaluation requires immutable Result/Summary evidence and coverage; packaging requires `READY`, an immutable image digest, manifest, and passing test report.
+- Do not register caller-computed evaluation results. Workers produce predictions, metrics, decisions, and summaries from frozen DatasetArtifact rowsets.
+- Keep credentials, registry secrets, Kubernetes details, and model binaries out of request JSON and reports.
+
+Report immutable input identities, returned run/job/package IDs, status, validation or gate failures, and the next authorized action. Never claim deployment; it is outside this Skill.

package/skills/model-lifecycle-management/agents/openai.yaml

@@ -0,0 +1,4 @@
+interface:
+  display_name: "模型训练、评估与打包"
+  short_description: "管理 ITSMP 模型训练、评估与不可变模型打包流程"
+  default_prompt: "使用 $model-lifecycle-management 管理模型训练、评估或打包流程。"

package/skills/model-lifecycle-management/references/evaluation.md

@@ -0,0 +1,33 @@
+# Evaluation Commands
+
+```bash
+ml-platform --profile server register-evaluation-config evaluation-config.json
+ml-platform --profile server register-metric-definition metric-definition.json
+ml-platform --profile server register-executable-package executable-package.json
+ml-platform --profile server verify-executable-package <name> <version>
+ml-platform --profile server publish-executable-package <name> <version>
+```
+
+Custom metric packages must be deterministic, digest-bound, network-restricted, verified, and separately authorized for publication.
+
+```bash
+ml-platform --profile server validate-evaluation-run evaluation-request.json
+ml-platform --profile server submit-evaluation evaluation-request.json
+ml-platform --profile server get-evaluation-run <run-id>
+ml-platform --profile server get-evaluation-job <job-id>
+ml-platform --profile server list-evaluation-attempts <job-id>
+ml-platform --profile server retry-evaluation-job <job-id>
+ml-platform --profile server cancel-evaluation-job <job-id>
+```
+
+An EvaluationRunRequest binds ModelArtifact(s), a frozen DatasetArtifact rowset, targets, EvaluationConfig, runtime identity, and optional baseline. Retry/cancel require separate authorization.
+
+```bash
+ml-platform --profile server get-evaluation-result <result-id>
+ml-platform --profile server get-evaluation-result-metrics <result-id>
+ml-platform --profile server get-evaluation-result-artifacts <result-id>
+ml-platform --profile server summarize-evaluation-run <run-id>
+ml-platform --profile server get-evaluation-summary <summary-id>
+```
+
+Report candidate, eligible, labeled and evaluable coverage; calculation space; slices; baseline identity; required decisions; runtime/package digests; and Result/Summary hashes. Exploratory or incomplete evidence is never a release gate pass.

package/skills/model-lifecycle-management/references/packaging.md

@@ -0,0 +1,17 @@
+# Model Packaging Commands
+
+```bash
+ml-platform --profile server validate-model-artifact <artifact-id>
+ml-platform --profile server create-model-package model-package-request.json
+ml-platform --profile server get-model-package <package-id> --package-version <version>
+```
+
+Confirm the ModelArtifact, signature, runtime, immutable base/builder/worker image digests, input modes, semantic package version, and destination repository. Creation submits a workload and needs separate authorization. Never put registry credentials in JSON.
+
+A deliverable package is `READY` and includes its remote image digest, immutable reference, manifest, signature and adapter hashes, runtime capabilities, and a passing test report for the same digest. Other states are not deliverable.
+
+```bash
+ml-platform --profile server cancel-model-package <package-id> --package-version <version>
+```
+
+Cancellation requires exact-package authorization. Do not rebuild an immutable failed version; correct inputs and create a new semantic package version. This Skill does not deploy images or write model-center delivery status.

package/skills/model-lifecycle-management/references/training.md

@@ -0,0 +1,33 @@
+# Training Commands
+
+Discover already-registered TrainerDefinitions (read-only) before authoring a
+TrainingRunSpec or deciding whether registration is needed:
+
+```bash
+ml-platform --profile server list-trainer-definitions
+ml-platform --profile server get-trainer-definition <name> <version> [--project default]
+```
+
+`get-trainer-definition` returns the entrypoint, immutable training image digest,
+`artifact_format`, `model_file`, capabilities, and parameter schema. Use it to confirm
+the referenced `v1` definition exists and its image digest matches the approved runtime
+release; discovery never mutates the registry. Registering a new definition is the
+separate mutation below, and a 404 means the definition (not the model plugin) is missing.
+
+```bash
+ml-platform --profile server register-trainer trainer-definition.json
+ml-platform --profile server validate-training-run training-run.json
+ml-platform --profile server register-training-run training-run.json
+ml-platform --profile server submit-training-job training-job-request.json
+ml-platform --profile server get-training-job <job-id>
+```
+
+The TrainingRun binds immutable DatasetArtifact identities, DataSchema, ordered inputs, targets, InputAdapter, TrainerDefinition `v1`, reproducibility controls, and runtime identity. Validation proves compatibility only. Submit the run and Job separately after authorization and record spec hash and Job ID. Terminal success must expose a ModelArtifact ID; validate it before evaluation or packaging.
+
+```bash
+ml-platform --profile server validate-model-artifact <artifact-id>
+ml-platform --profile server retry-training-job <job-id>
+ml-platform --profile server cancel-training-job <job-id>
+```
+
+Retry and cancellation require exact-Job authorization. Never change a run contract while retrying; create a new immutable run when semantic inputs change.